Intelligent power generation control system optimization method and system

Abstract

The invention discloses an intelligent power generation control system optimization method and system, and relates to the technical field of power generation control system optimization, and the method comprises the steps: defining the working condition situations of a power generation control system according to the working condition related information, formulating the judgment standard of each working condition situation, and carrying out the comprehensive and unified determination of the working condition situations, and a reliable basis is provided for dynamic prediction of a subsequent sliding pressure curve. And a working condition prediction model of the power generation control system is constructed by combining operation task information, so that the working condition is predicted in advance and the accuracy of working condition identification is ensured. The sliding pressure prediction window length is set for each working condition situation in the working condition situation change content to predict the sliding pressure curve, the sliding pressure curve parameters are adjusted in advance, passive correction is avoided, and the accuracy of sliding pressure curve prediction is improved. The power generation control system is optimized in multiple aspects by means of deviation information, and comprehensive control optimization in the aspects of safety, efficiency, economy and the like is carried out on the power generation control system on the basis of a flexible high-precision predicted sliding pressure curve.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power generation control system optimization, and particularly relates to an intelligent power generation control system optimization method and system. BACKGROUND

[0002] In the intelligent power generation control system, the sliding pressure curve optimization is the key technology to improve the economic efficiency and flexibility of the unit operation. The traditional sliding pressure operation adjusts the main steam pressure to adapt to the load change, but the fixed curve is difficult to balance the optimal heat rate under different working conditions. For example, when the thermal power unit is at low load, the throttling of the high-pressure regulating valve leads to efficiency decline, and although the sliding pressure operation can reduce the throttling loss, the reduction of the main steam pressure will sacrifice the cycle efficiency, and the best sliding pressure point needs to be determined by the test method to balance the two. With the expansion of unit capacity and the improvement of parameters, the traditional DCS architecture is difficult to realize the dynamic optimization of the sliding pressure curve due to the limitation of processing capacity, especially under the background of increasing proportion of new energy, the unit needs to participate in the frequency modulation and other fast load change scenes frequently, which puts higher requirements on the real-time correction ability of the sliding pressure curve. The intelligent control technology can dynamically adjust the sliding pressure set value by introducing big data analysis and machine learning algorithm, such as correcting the back pressure influence based on the condenser pressure, or realizing the valve point given correction through closed-loop PID control, so as to maintain the optimal economic valve position in the whole working condition range, and significantly improve the unit response speed and energy utilization efficiency.

[0003] In the prior art, because the fixed sliding pressure curve cannot adapt to the dynamic demand of the complex and changeable power generation control system, the accuracy and adaptability of the power generation control system optimization are poor, and the safe and efficient operation of the power generation control system cannot be guaranteed.

[0004] Therefore, how to improve the accuracy and adaptability of the power generation control system optimization is a technical problem to be solved at present. SUMMARY

[0005] The purpose of the present application is to solve the problem in the prior art that the fixed sliding pressure curve cannot adapt to the dynamic demand of the complex and changeable power generation control system, and the accuracy and adaptability of the power generation control system optimization are poor, and an intelligent power generation control system optimization method is proposed, which comprises,

[0006] Obtaining the operation task information and working condition related information of the power generation control system, defining the working condition context of the power generation control system according to the working condition related information, and constructing the working condition prediction model of the power generation control system in combination with the operation task information;

[0007] Predicting the working condition context change content of the power generation control system in a future period of time through the working condition prediction model of the power generation control system, obtaining the sliding pressure related information of the power generation control system, and setting the sliding pressure prediction window length for each working condition context in the working condition context change content.

[0008] The sliding pressure curve prediction model is used to predict the sliding pressure curve of each working condition scenario in the working condition scenario change content through the sliding pressure prediction window length and the sliding pressure curve prediction model;

[0009] On the basis of the predicted sliding pressure curve, deviation information is confirmed, and the power generation control system is optimized in multiple aspects by means of the deviation information.

[0010] In some embodiments of the present application, the working condition scenarios of the power generation control system are defined according to working condition related information, including,

[0011] The working condition scenarios of the power generation control system include single working condition scenarios and transition working condition scenarios;

[0012] The working condition key parameters of all working condition scenarios under the single working condition scenarios and the transition working condition scenarios are screened from the working condition related information, and a matching relationship between the working condition scenarios and the working condition key parameters is established;

[0013] The historical values of the working condition key parameters are intercepted according to the matching relationship between the working condition scenarios and the working condition key parameters, the characteristic values of the historical values of each key parameter are counted, and the initial intervals of the key parameters of the single working condition scenarios and the transition working condition scenarios are determined according to the characteristic values;

[0014] The aging index and the environmental interference index of the power generation control system are determined, the initial intervals of the key parameters of the single working condition scenarios and the transition working condition scenarios are adjusted through the aging index and the environmental interference index, and the target intervals of the key parameters of the single working condition scenarios and the transition working condition scenarios are obtained.

[0015] In some embodiments of the present application, the initial intervals of the key parameters of the single working condition scenarios and the transition working condition scenarios are adjusted through the aging index and the environmental interference index, including,

[0016] The threshold values of the aging index and the environmental interference index are set, and the high aging, the low aging, the high interference and the low interference are distinguished by means of the threshold values;

[0017] If the aging index is high aging and the environmental interference index is high interference, or the aging index is low aging and the environmental interference index is low interference, the aging index and the environmental interference index are combined through a first combination mode to determine a first aging interference index;

[0018] If the aging index is low aging and the environmental interference index is high interference, or the aging index is high aging and the environmental interference index is low interference, the aging index and the environmental interference index are combined through a second combination mode to determine a second aging interference index;

[0019] Adjust initial intervals of key parameters of the single-working-condition context and the transition-working-condition context by the first aging interference index or the second aging interference index.

[0020] In some embodiments of the present application, a working condition prediction model of the power generation control system is constructed in combination with the running task information, including,

[0021] The running task information and the working condition related information corresponding to all working condition contexts of the two types of single-working-condition context and transition-working-condition context are intercepted from the previous running task information and the working condition related information of the power generation control system, the running task instructions and the key parameter features are extracted, and a sample set of the running task instructions and the key parameter features is formed;

[0022] For the single-working-condition context, the matching degree between the running task instructions and the specific type under the single-working-condition context is calculated, denoted as task matching degree, the matching degree between the key parameter features and the target interval of the key parameter is calculated, denoted as parameter matching degree, the comprehensive matching degree of the single-working-condition context is determined based on the task matching degree and the parameter matching degree, and the comprehensive matching degree of the single-working-condition context is marked as a label on the sample set;

[0023] For the transition-working-condition context, the target interval of the key parameter of the transition-working-condition context includes the parameter interval and the parameter change rate interval, the matching degree between the key parameter features and the parameter change rate interval is calculated, denoted as dynamic matching degree, the comprehensive matching degree of the transition-working-condition context is determined based on the task matching degree, the parameter matching degree and the dynamic matching degree, and the comprehensive matching degree of the transition-working-condition context is marked as a label on the sample set;

[0024] The working condition prediction model of the power generation control system is trained according to the sample sets and the labels of the single-working-condition context and the transition-working-condition context.

[0025] In some embodiments of the present application, a sliding pressure prediction window length is set for each working condition context in the working condition context change content, including,

[0026] The stability degree of the key parameter of each working condition context is extracted according to the matching relationship of the working condition context-key parameter;

[0027] The working condition context is matched with the sliding pressure related information, and the stability degree of the sliding pressure dynamic response parameter and the sliding pressure control target of each working condition context are extracted;

[0028] For the same working condition context, the sliding pressure prediction window length is set based on the stability degree of the key parameter, the stability degree of the sliding pressure dynamic response parameter and the sliding pressure control target.

[0029] In some embodiments of the present application, a sliding pressure curve prediction model is established, including,

[0030] The sliding pressure curve prediction model adopts an LSTM-Attention hybrid network structure, and comprises an input layer, an LSTM layer, an Attention layer and a full connection layer, and adopts a weighted MAE loss function.

[0031] According to the feature hierarchical sampling according to the working condition context, the sliding pressure curve prediction model is trained and optimized.

[0032] In some embodiments of the application, deviation information is confirmed on the basis of the predicted sliding pressure curve, including,

[0033] The actual sliding pressure curve is compared with the predicted sliding pressure curve in multiple dimensions to generate deviation information, and the deviation information includes time sequence deviation, amplitude deviation, trend deviation and working condition adaptation deviation.

[0034] In some embodiments of the application, the deviation information is used to optimize the power generation control system in multiple aspects, including,

[0035] The time sequence deviation, the amplitude deviation, the trend deviation and the working condition adaptation deviation are comprehensively used to optimize the control strategy of the power generation control system.

[0036] Correspondingly, an intelligent power generation control system optimization system comprises,

[0037] A first module is configured to acquire running task information and working condition related information of a power generation control system, define working condition contexts of the power generation control system according to the working condition related information, and construct a working condition prediction model of the power generation control system in combination with the running task information;

[0038] A second module is configured to predict working condition context change content of the power generation control system in a future period of time through the working condition prediction model of the power generation control system, acquire sliding pressure related information of the power generation control system, and set a sliding pressure prediction window length for each working condition context in the working condition context change content.

[0039] A third module is configured to establish a sliding pressure curve prediction model, and predict a sliding pressure curve for each working condition context in the working condition context change content through the sliding pressure prediction window length and the sliding pressure curve prediction model.

[0040] A fourth module is configured to confirm deviation information on the basis of the predicted sliding pressure curve, and use the deviation information to optimize the power generation control system in multiple aspects.

[0041] Compared with the prior art, the application has the following beneficial effects:

[0042] 1. Define the working condition context of the power generation control system according to the working condition related information, formulate the judgment standard of each working condition context, comprehensively and uniformly identify the working condition context, and provide a reliable basis for the subsequent dynamic prediction of the sliding pressure curve. The working condition prediction model of the power generation control system is constructed combined with the operation task information, the working condition is predicted in advance, the accuracy of working condition identification is ensured, and the basis for setting the subsequent sliding pressure prediction window length is provided.

[0043] 2. The sliding pressure prediction window length is set for each working condition in the working condition context change content to predict the sliding pressure curve, the sliding pressure curve parameters are adjusted in advance to avoid passive correction, and the accuracy of the sliding pressure curve prediction is improved. The deviation information is used to optimize the power generation control system in multiple aspects, and the sliding pressure curve is predicted with flexible high precision as the basis, and the power generation control system is comprehensively controlled and optimized in safety, efficiency and economy. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 A flowchart of an intelligent power generation control system optimization method is provided for the present application.

[0045] Figure 2 A structure diagram of an intelligent power generation control system optimization system is provided for the present application. DETAILED DESCRIPTION

[0046] The technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all embodiments.

[0047] Reference Figure 1 An intelligent power generation control system optimization method, comprising the following steps:

[0048] Step S101, obtain the operation task information and working condition related information of the power generation control system, define the working condition context of the power generation control system according to the working condition related information, and construct the working condition prediction model of the power generation control system combined with the operation task information.

[0049] In this embodiment, the working condition context includes steady state operation, climbing, deep peak regulation, fault precursor, start-stop machine and the like, and the working condition related information includes main steam pressure, load rate, feed water temperature, exhaust gas temperature, load change rate, main steam pressure change rate, turbine valve opening change rate and the like.

[0050] In some embodiments of the present application, the working condition context of the power generation control system is defined according to the working condition related information, which includes,

[0051] The working condition context of the power generation control system includes single working condition context and transition working condition context;

[0052] Screening the working condition key parameters of all working condition contexts under single working condition context and transition working condition context from the working condition related information, and establishing the matching relationship of working condition context-working condition key parameters;

[0053] According to the matching relationship of working condition context-working condition key parameters, intercepting the historical values of working condition key parameters, and counting the characteristic values of historical values of each key parameter, and determining the initial interval of key parameters of single working condition context and transition working condition context according to the characteristic values;

[0054] Determine the aging index and environmental interference index of the power generation control system, and adjust the initial interval of key parameters of single working condition context and transition working condition context through the aging index and environmental interference index, to obtain the target interval of key parameters of single working condition context and transition working condition context.

[0055] In this embodiment, the working condition of the power generation control system is divided into single working condition context (such as steady state full load, steady state low load, etc.) and transition working condition context (such as load climbing, load sliding, start-stop transition, etc.), which is a partial list, and the working condition can be further limited according to the actual situation. Single working condition context is the steady state of a working condition, and transition working condition context is the process of transition from one working condition to another. A "working condition context-key parameter" mapping table is constructed, for example:

[0056] Single working condition: main steam pressure, load rate, feed water temperature, exhaust gas temperature;

[0057] Transition working condition: load change rate, main steam pressure change rate, turbine governing valve opening change rate.

[0058] According to the matching relationship, intercept the historical data of the past one year, count the characteristic values of each parameter, calculate the mean, standard deviation, extreme value, etc., and consider the characteristic values to determine the initial interval of key parameters of single working condition context and transition working condition context.

[0059] Aging index: calculated based on the length of equipment operation and maintenance records. Performance degradation model: establish a mathematical model of the change of equipment performance with time, for example, use the inverse aging model. By measuring the initial value and current value of the key performance indicators of the equipment (such as vibration frequency, temperature stability, etc.), the aging index is calculated.

[0060] Interference index: calculated based on environmental temperature fluctuations, ambient noise, electromagnetic interference, etc. Noise spectrum analysis quantifies ambient noise using equivalent continuous sound level (Leq) or noise curves (such as NC, NR curves). For example, by measuring the complete sound spectrum of an empty room, the measured noise spectrum is superimposed within the equivalent curve family, and the lowest noise curve corresponding to the measured noise spectrum is the grade of the corresponding spectrum. Vibration interference quantification uses acceleration sensors to measure equipment vibration amplitude and combines spectral analysis to determine the main interference frequency. Electromagnetic interference quantification measures the electromagnetic interference strength on the power supply line using a spectrum analyzer to assess its impact on device performance. The interference index is determined by combining several aspects.

[0061] Adjust the initial interval boundaries of the key parameters of the single working condition and the transition working condition by combining the aging situation and the environmental interference situation.

[0062] In some embodiments of the present application, the initial intervals of the key parameters of the single working condition and the transition working condition are adjusted by the aging index and the environmental interference index, including,

[0063] Set threshold values for the aging index and the environmental interference index, and use the threshold values to distinguish between high aging, low aging, high interference, and low interference.

[0064] If the aging index is high aging and the environmental interference index is high interference, or the aging index is low aging and the environmental interference index is low interference, combine the aging index and the environmental interference index by a first combination method to determine a first aging interference index.

[0065] If the aging index is low aging and the environmental interference index is high interference, or the aging index is high aging and the environmental interference index is low interference, combine the aging index and the environmental interference index by a second combination method to determine a second aging interference index.

[0066] Adjust the initial intervals of the key parameters of the single working condition and the transition working condition by the first aging interference index or the second aging interference index.

[0067] In the present embodiment, the first combination method is applicable when the aging and interference levels are in the same direction (same high or same low), and the weighted geometric mean method gives the same weight to interference (D) and aging (A) to calculate the aging interference comprehensive index.

[0068]

[0069] where the first combination method is the weighted geometric mean, when the interference and aging trends are consistent, the effects of both on device performance are synergistic, and the comprehensive impact assessment needs to be strengthened, EDI is the first aging interference index, D is the aging index, and A is the environmental interference index.

[0070]

[0071] wherein the second combination mode is a dynamic weight distribution weighted average, when the interference and the aging trend are opposite, the weight needs to be dynamically adjusted to reflect the dominant factor, EDI2 is a second aging interference index, a1 (D-D0) is a combination weight determined by the difference between the aging index and the aging index threshold, a2 (A-A0) is a combination weight determined by the difference between the environmental interference index and the environmental interference index threshold, D is the aging index, and A is the environmental interference index.

[0072] The first aging interference index or the second aging interference index determines a correction factor through different mapping relationships, the correction factor is used to adjust two endpoints of an interval (in a product form), and the distribution relationship of the correction factors corresponding to different key parameters is different, that is, the determined overall correction factor is distributed to each key parameter for specific correction factor distribution.

[0073] It should be noted that the target interval of the key parameter of the single working condition context refers to the interval of the specific parameter, and the target interval of the key parameter of the transition working condition context refers to the interval of the specific parameter and the interval of the parameter change.

[0074] In some embodiments of the present application, a working condition prediction model of a power generation control system is constructed in combination with running task information, including,

[0075] The running task information and the working condition related information corresponding to all working condition contexts of the two types of single working condition contexts and transition working condition contexts are intercepted from the past running task information and the working condition related information of the power generation control system, the running task instructions and the key parameter features are extracted, and a sample set of the running task instructions and the key parameter features is formed;

[0076] For the single working condition context, the matching degree between the running task instruction and the specific type under the single working condition context is calculated, denoted as a task matching degree, the matching degree between the key parameter feature and the target interval of the key parameter is calculated, denoted as a parameter matching degree, the comprehensive matching degree of the single working condition context is determined based on the task matching degree and the parameter matching degree, and the comprehensive matching degree of the single working condition context is marked as a label on the sample set;

[0077] For the transition working condition context, the target interval of the key parameter of the transition working condition context includes the parameter interval and the parameter change rate interval, the matching degree between the key parameter feature and the parameter change rate interval is calculated, denoted as a dynamic matching degree, the comprehensive matching degree of the transition working condition context is determined based on the task matching degree, the parameter matching degree and the dynamic matching degree, and the comprehensive matching degree of the transition working condition context is marked as a label on the sample set;

[0078] The working condition prediction model of the power generation control system is trained according to the sample sets and the labels of the single working condition context and the transition working condition context.

[0079] In this embodiment, the running task information includes (such as load instruction, fuel instruction, etc.), static characteristics: mean, standard deviation, extreme value (such as main steam pressure 16.25±0.5MPa). Dynamic characteristics: change rate, fluctuation frequency (such as pressure change rate ≤0.1MPa / min), mainly the transition condition changes greatly, which needs dynamic characteristics. Task matching degree, calculate the similarity of running task instruction and preset working condition type (such as "full load operation"). Parameter matching degree, calculate through the Euclidean distance, Manhattan distance, etc. between parameters and target interval, if the parameter is located in the target interval, it is 0. Dynamic matching degree, calculate the matching degree of parameter change rate and target interval, which can be described by Jaccard coefficient. Based on task matching degree and parameter matching degree, and based on task matching degree, parameter matching degree and dynamic matching degree, respectively determine the comprehensive matching degree of single working condition context and transition working condition context respectively, and mark on the sample. The comprehensive matching degree (0-1 continuous value) quantifies the degree of fit between the working condition and the task and the parameter, provides a more fine-grained supervision signal, gives higher weight to the samples with low comprehensive matching degree (such as fault working condition), improves the adaptability of the model to extreme conditions, and improves the prediction accuracy of the model.

[0080] In step S102, the working condition prediction model of the power generation control system is used to predict the working condition context change content of the power generation control system in a future period of time, obtain the sliding pressure related information of the power generation control system, and set a sliding pressure prediction window length for each working condition context in the working condition context change content.

[0081] In this embodiment, the running task information is input into the working condition prediction model to predict the working condition context change content in a future period of time. The self-attention mechanism of Transformer is used to capture the long-time dependence and multi-parameter interaction of working condition parameters, and the working condition type and parameter change are predicted through the context classification head and the regression head respectively.

[0082] Transformer encoder:

[0083] Multi-head attention mechanism: capture the space-time correlation between parameters (such as the causal relationship between load instruction and main steam pressure);

[0084] Position encoding: add time sequence information to time series data;

[0085] Layer normalization and residual connection: stabilize the training process.

[0086] Context classification and regression head:

[0087] Classification head: fully connected layer + Softmax, output future working condition type probability (such as "steady full load" "30% load increase");

[0088] Regression head: fully connected layer, output the predicted value sequence of future key parameters (such as the value of main steam pressure every minute in the next 10 minutes) and dynamic matching degree (parameter change rate compliance score).

[0089] The generator (LSTM) generates the future working condition parameter sequence, and the discriminator (CNN) judges the authenticity of the sequence to realize end-to-end working condition evolution prediction.

[0090] In some embodiments of the application, a sliding pressure prediction window length is set for each working condition context in the working condition context change content, including,

[0091] According to the matching relationship between the working condition context and the key parameters of the working condition, the stability degree of the key parameters of each working condition context is extracted;

[0092] The working condition context is matched with the sliding pressure related information, and the stability degree of the sliding pressure dynamic response parameter and the sliding pressure control target of each working condition context are extracted;

[0093] For the same working condition context, the sliding pressure prediction window length is set based on the stability degree of the key parameters, the stability degree of the sliding pressure dynamic response parameter and the sliding pressure control target.

[0094] In this embodiment,

[0095] Among them, is the sliding pressure prediction window length of the ith1 working condition context, is the prediction conversion coefficient of the ith1 working condition context, n1 and n2 are the number of key parameters and sliding pressure dynamic response parameters, are the stability weights corresponding to the ith2 key parameter and the ith3 sliding pressure dynamic response parameter respectively, are the stability degrees corresponding to the ith2 key parameter and the ith3 sliding pressure dynamic response parameter of the ith1 working condition context (obtained by analyzing the standard deviation, coefficient of variation of the key parameter and the meaning of the sliding pressure dynamic response parameter), is the first constant of the ith1 working condition context, is the sliding pressure control target of the ith1 working condition context, is the initial sliding pressure prediction window length of the ith1 working condition context, represents the initial sliding pressure prediction window length obtained by mapping the sliding pressure control target, represents the correction of the stability of the sliding pressure dynamic response to the stability of the key parameter, and the first constant is to balance the size of the correction function, is the class average, which is slightly larger than the average, so that the comprehensive stability is more reliable.

[0096] Key parameter stability: reflects the stability of the operating condition context itself (e.g., the fluctuation of the main steam pressure in steady-state operation).

[0097] Evaluation index: standard deviation (σ), coefficient of variation (CV = σ / mean), range (R), and other statistical quantities.

[0098] Physical meaning: the smaller σ / CV, the smoother the parameter fluctuation, and the more stable the operating condition.

[0099] Sliding pressure dynamic response parameter stability: represents the dynamic adjustment capability of the parameter in the sliding pressure control process (e.g., the rate of change of the turbine governing valve opening during load increase).

[0100] Evaluation index: dynamic response time (τ), overshoot (M), steady-state error (e_ss), and other control theory indicators.

[0101] Physical meaning: the shorter τ and the smaller M, the more stable the dynamic response, and the more accurate the sliding pressure control.

[0102] Sliding pressure control target type (partially listed, different targets correspond to different initial sliding pressure prediction window lengths):

[0103] Economic target: minimize turbine heat consumption (e.g., sliding pressure curve close to design value);

[0104] Safety target: avoid parameter out-of-limit (e.g., main steam pressure does not exceed the safety valve take-off value);

[0105] Flexibility target: fast response to load instruction (e.g., sliding pressure adjustment speed in frequency modulation scenario).

[0106] Based on the stability of the key parameters, the stability of the sliding pressure dynamic response parameters, and the sliding pressure control target, the sliding pressure prediction window length is set to ensure adaptation to the operating condition context, so that the timeliness and accuracy of the sliding pressure curve prediction are improved.

[0107] Step S103, establish a sliding pressure curve prediction model, and predict the sliding pressure curve for each operating condition context in the operating condition context change content through the sliding pressure prediction window length and the sliding pressure curve prediction model.

[0108] In some embodiments of the present application, the sliding pressure curve prediction model is established, including,

[0109] The sliding pressure curve prediction model adopts an LSTM-Attention hybrid network structure, and includes an input layer, an LSTM layer, an Attention layer, and a fully connected layer, and uses a weighted MAE loss function.

[0110] According to the feature hierarchical sampling of the operating condition context, the sliding pressure curve prediction model is trained and optimized.

[0111] In this embodiment, the LSTM-Attention hybrid network is designed as follows:

[0112] (1) Network structure hierarchy

[0113] Input layer:

[0114] Input features: including key parameters (e.g., main steam pressure, feedwater flow), dynamic response parameters (e.g., valve opening rate of change), sliding pressure control targets (e.g., "heat consumption optimization" encoding), and dynamic window length (from step S102).

[0115] Data preprocessing: normalize (Min-Max) and sequence alignment (sliding window clipping) for multi-source heterogeneous data (time series signal + discrete target).

[0116] LSTM layer:

[0117] Function: capture the time series dependence of sliding pressure curve (e.g., pressure lag effect with load change).

[0118] Configuration: 2-layer bidirectional LSTM, 128 hidden units per layer, ReLU activation function to suppress gradient disappearance.

[0119] Input: preprocessed time series features (e.g., past 60 seconds of parameter sampling values).

[0120] Attention layer:

[0121] Function: strengthen the weight of key time steps (e.g., load mutation points) and suppress noise interference.

[0122] Implementation: use scaled dot-product attention (Scaled Dot-Product Attention) to calculate the weighted sum of Query-Key-Value.

[0123] Fully connected layer:

[0124] Output: sliding pressure curve prediction value for future Ti seconds (dynamic window length) (e.g., main steam pressure prediction sequence).

[0125] Activation function: linear output (regression task) with Dropout (0.2) to prevent overfitting.

[0126] (2) Weighted MAE loss function

[0127] Design motivation: traditional MAE treats all time step errors equally, while in sliding pressure control, recent errors (e.g., future 5 seconds) have a much greater impact on control instruction generation than long-term errors.

[0128] Feature hierarchical sampling

[0129] Hierarchical logic:

[0130] Working condition context layer: divide datasets by load command type (e.g. steady state / load increase / load decrease);

[0131] Parameter feature layer: densely sample key parameters (high weight) (e.g. 1 sample per second), sparsely sample auxiliary parameters (low weight) (e.g. 1 sample per 5 seconds).

[0132] Data augmentation:

[0133] Time warping: randomly stretch / compress time series segments (DTW algorithm), improve model robustness to dynamic changes;

[0134] Noise injection: add Gaussian noise (σ = 0.01) to training data, simulate actual sensor errors.

[0135] Model optimization process

[0136] Initialization:

[0137] Use He initialization for weights, set LSTM layer forget gate bias to 1 (mitigate gradient vanishing);

[0138] Learning rate warm-up (linearly increase to 10-3 for the first 5 epochs).

[0139] Training phase:

[0140] Batch division: stratified sampling by working condition context, each batch contains samples of similar context (e.g. all "load increase 30%" working conditions);

[0141] Gradient clipping: limit LSTM gradient norm (clip = 1.0) to avoid exploding gradients.

[0142] Verification and parameter tuning:

[0143] Use TimeSeriesSplit to reserve 20% of consecutive data as the validation set;

[0144] Hyperparameter search (Optuna): optimize LSTM layer number (1-3 layers), attention head number (2-8 heads) and λ value (1.0-1.5).

[0145] Step S104, on the basis of the predicted sliding pressure curve, the deviation information is confirmed, and the power generation control system is optimized in multiple aspects by means of the deviation information.

[0146] In some embodiments of the present application, on the basis of the predicted sliding pressure curve, the deviation information is confirmed, including,

[0147] The actual sliding pressure curve and the predicted sliding pressure curve are compared in multiple dimensions to generate deviation information, including time sequence deviation, amplitude deviation, trend deviation, and working condition adaptation deviation.

[0148] In some embodiments of the present application, the deviation information is used to optimize the power generation control system in multiple aspects, including,

[0149] The control strategy of the power generation control system is optimized by comprehensively considering the time sequence deviation, amplitude deviation, trend deviation, and working condition adaptation deviation.

[0150] In the present embodiment, the predicted sliding pressure curve (p^(t)) and the actual sliding pressure curve (p(t)) are aligned by timestamp, and the dynamic time warping (DTW) algorithm is used to handle the problem of inconsistent sampling frequency or phase shift, thereby realizing the time alignment of the two sliding pressure curves. The time sequence deviation is the degree of misalignment of the predicted and actual curves on the time axis. The amplitude deviation is the absolute error between the predicted value and the actual value. The trend deviation is the difference in slope between the predicted curve and the actual curve. The working condition adaptation deviation is the distribution difference between the current working condition and the training working condition. For example, adaptive adjustment of control parameters, optimization of PID controller, and dynamic adjustment of PID parameters according to amplitude deviation and trend deviation.

[0151] Online learning mechanism: when the working condition adaptation deviation is triggered, incremental learning is started:

[0152] Collect new data (such as the last 100 time steps) from the current working condition;

[0153] Freeze the Attention layer of the LSTM-Attention model, and only fine-tune the LSTM layer weights;

[0154] Update the model using small batch gradient descent (batch size = 32).

[0155] Control strategy switching:

[0156] If the working condition adaptation deviation of the last 5 time steps exceeds the threshold value, automatically switch to a rule-based control strategy (such as lookup table method), and record the abnormal working condition data for subsequent model optimization.

[0157] Correspondingly, an intelligent power generation control system optimization system, as shown in Figure 2 includes,

[0158] The first module is used to obtain the running task information and working condition related information of the power generation control system, define the working condition context of the power generation control system according to the working condition related information, and construct a working condition prediction model of the power generation control system in combination with the running task information;

[0159] The second module is configured to predict the working condition situation change content of the power generation control system in a future period of time by a working condition prediction model of the power generation control system, acquire the sliding pressure related information of the power generation control system, and set a sliding pressure prediction window length for each working condition situation in the working condition situation change content.

[0160] The third module is configured to establish a sliding pressure curve prediction model, and predict the sliding pressure curve for each working condition situation in the working condition situation change content by the sliding pressure prediction window length and the sliding pressure curve prediction model.

[0161] The fourth module is configured to confirm the deviation information on the basis of the predicted sliding pressure curve, and optimize the power generation control system in multiple aspects by means of the deviation information.

[0162] Compared with the prior art, the present application has the following beneficial effects:

[0163] 1. The working condition situation of the power generation control system is defined according to the working condition related information, the determination standard of each working condition situation is formulated, the working condition situation is comprehensively and uniformly identified, a reliable basis is provided for the subsequent dynamic prediction of the sliding pressure curve, the working condition prediction model of the power generation control system is constructed in combination with the operation task information, the working condition is predicted in advance, the accuracy of the working condition identification is ensured, and a basis is provided for the subsequent setting of the sliding pressure prediction window length.

[0164] 2. The sliding pressure prediction window length is set for each working condition situation in the working condition situation change content to predict the sliding pressure curve, the sliding pressure curve parameters are adjusted in advance, passive correction is avoided, the accuracy of the sliding pressure curve prediction is improved, the deviation information is used to optimize the power generation control system in multiple aspects, and the power generation control system is comprehensively controlled and optimized in safety, efficiency and economy based on the flexible and high-precision predicted sliding pressure curve.

[0165] Through the description of the above embodiments, those skilled in the art can clearly understand that the present application can be implemented by hardware, or by means of software and a necessary general hardware platform. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a U disk, a mobile hard disk, etc.), and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various implementation scenarios of the present application.

[0166] Those skilled in the art can understand that the drawings are only a schematic diagram of a preferred embodiment, and the modules or flows in the drawings are not necessarily required for implementing the present application.

[0167] Those skilled in the art can understand that the modules in the system in the implementation scenario can be distributed in the system in the implementation scenario according to the description of the implementation scenario, or can be changed to be located in one or more systems different from the implementation scenario. The modules in the above implementation scenario can be combined into one module, or can be further split into multiple sub-modules.

[0168] The above merely describes the preferred embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can make equivalent replacements or changes within the technical range disclosed by the present application according to the technical solutions and the inventive concept of the present application, which should be covered within the protection scope of the present application.

Claims

1. An optimization method for an intelligent power generation control system, characterized in that, The method comprises the following steps: obtaining operation task information and working condition related information of the power generation control system, defining working condition scenarios of the power generation control system according to the working condition related information, and constructing a working condition prediction model of the power generation control system in combination with the operation task information; predicting the working condition scenario change content of the power generation control system in a future period of time through the working condition prediction model of the power generation control system, obtaining the sliding pressure related information of the power generation control system, and setting a sliding pressure prediction window length for each working condition scenario in the working condition scenario change content; establishing a sliding pressure curve prediction model, and predicting the sliding pressure curve of each working condition scenario in the working condition scenario change content through the sliding pressure prediction window length and the sliding pressure curve prediction model; on the basis of the predicted sliding pressure curve, confirming deviation information, and optimizing the power generation control system in multiple aspects by means of the deviation information.

2. The method of claim 1, wherein, The working condition scenarios of the power generation control system are defined according to the working condition related information, which comprises the following steps: The working condition scenarios of the power generation control system include single working condition scenarios and transition working condition scenarios; all working condition key parameters of the single working condition scenarios and the transition working condition scenarios are screened out from the working condition related information, and a matching relationship between the working condition scenarios and the working condition key parameters is established; the historical values of the working condition key parameters are intercepted according to the matching relationship between the working condition scenarios and the working condition key parameters, the characteristic values of the historical values of each key parameter are counted, and the initial intervals of the key parameters of the single working condition scenarios and the transition working condition scenarios are determined according to the characteristic values; the aging index and the environmental interference index of the power generation control system are determined, the initial intervals of the key parameters of the single working condition scenarios and the transition working condition scenarios are adjusted through the aging index and the environmental interference index, and the target intervals of the key parameters of the single working condition scenarios and the transition working condition scenarios are obtained.

3. The method of claim 2, wherein, The initial intervals of the key parameters of the single working condition scenarios and the transition working condition scenarios are adjusted through the aging index and the environmental interference index, which comprises the following steps: The aging index and the environmental interference index are set with respective threshold values, and the threshold values are used to distinguish high aging, low aging, high interference and low interference; if the aging index is high aging and the environmental interference index is high interference, or the aging index is low aging and the environmental interference index is low interference, the aging index and the environmental interference index are combined through a first combination mode to determine a first aging interference index; if the aging index is low aging and the environmental interference index is high interference, or the aging index is high aging and the environmental interference index is low interference, the aging index and the environmental interference index are combined through a second combination mode to determine a second aging interference index; the initial intervals of the key parameters of the single working condition scenarios and the transition working condition scenarios are adjusted through the first aging interference index or the second aging interference index. The working condition prediction model of the power generation control system is constructed in combination with the operation task information, which comprises the following steps:

4. The method of claim 2, wherein, the operation task information and the working condition related information corresponding to all working condition scenarios under the single working condition scenarios and the transition working condition scenarios are intercepted from the past operation task information and the working condition related information of the power generation control system, the operation task instructions and the key parameter characteristics are extracted, and a sample set of the operation task instructions and the key parameter characteristics is formed; ​ For a single working condition context, a matching degree between the running task instruction and the specific type under the single working condition context is calculated, denoted as a task matching degree, a matching degree between the key parameter feature and the target interval of the key parameter is calculated, denoted as a parameter matching degree, a comprehensive matching degree of the single working condition context is determined based on the task matching degree and the parameter matching degree, and the comprehensive matching degree of the single working condition context is marked as a label on the sample set; For a transition working condition context, the target interval of the key parameter of the transition working condition context includes a parameter interval and a parameter change rate interval, a matching degree between the key parameter feature and the parameter change rate interval is calculated, denoted as a dynamic matching degree, a comprehensive matching degree of the transition working condition context is determined based on the task matching degree, the parameter matching degree and the dynamic matching degree, and the comprehensive matching degree of the transition working condition context is marked as a label on the sample set; The working condition prediction model of the power generation control system is trained according to the sample sets and the labels of the single working condition context and the transition working condition context.

5. The method of claim 2, wherein, For each working condition context in the working condition context change content, a sliding pressure prediction window length is set, including, According to the matching relationship of the key parameters of the working condition context-working condition, the stability degree of the key parameters of each working condition context is extracted; The working condition context is matched with the sliding pressure related information, and the stability degree of the sliding pressure dynamic response parameter and the sliding pressure control target of each working condition context are extracted; For the same working condition context, the stability degree of the key parameter, the stability degree of the sliding pressure dynamic response parameter and the sliding pressure control target are used to set the sliding pressure prediction window length.

6. The method of claim 2, wherein, A sliding pressure curve prediction model is established, including, The sliding pressure curve prediction model adopts an LSTM-Attention hybrid network structure, the sliding pressure curve prediction model includes an input layer, an LSTM layer, an Attention layer and a full connection layer, and a weighted MAE loss function is used; According to the working condition context, the feature is hierarchically sampled, and the sliding pressure curve prediction model is trained and optimized.

7. The method of claim 2, wherein, On the basis of the predicted sliding pressure curve, deviation information is confirmed, including, The actual sliding pressure curve is compared with the predicted sliding pressure curve in multiple dimensions to generate deviation information, and the deviation information includes time sequence deviation, amplitude deviation, trend deviation and working condition adaptation deviation.

8. The method of claim 2, wherein, The deviation information is used to optimize the power generation control system in multiple aspects, including, The time sequence deviation, the amplitude deviation, the trend deviation and the working condition adaptation deviation are used to optimize the control strategy of the power generation control system.

9. An intelligent generation control system optimization system, characterized by, including, The first module is used to acquire the running task information and the working condition related information of the power generation control system, define the working condition context of the power generation control system according to the working condition related information, and construct the working condition prediction model of the power generation control system in combination with the running task information; The second module is used to predict the working condition context change content of the power generation control system in a future period of time through the working condition prediction model of the power generation control system, acquire the sliding pressure related information of the power generation control system, and set the sliding pressure prediction window length for each working condition context in the working condition context change content; The third module is used to establish a sliding pressure curve prediction model, and predict the sliding pressure curve of each working condition context in the working condition context change content through the sliding pressure prediction window length and the sliding pressure curve prediction model. The fourth module is used for confirming deviation information on the basis of the predicted sliding pressure curve, and the power generation control system is optimized in multiple aspects by means of the deviation information.

Images