Data mining prediction method for combined cycle unit coupled with multi-source energy storage

Abstract

The application discloses a data mining prediction method for combined cycle unit coupling multi-source energy storage, and particularly relates to the technical field of data mining prediction, and through the construction of an association rule library and a response surface model, wide-area operation data are collected to generate a benchmark energy efficiency time sequence vector sequence and a boundary condition time sequence matrix, a time sequence lag correlation degree coefficient between energy efficiency fluctuation and boundary parameter fluctuation is calculated, and a significant lag effect period is marked; a dynamic adjustment correction factor is calculated based on the boundary condition, a benchmark energy efficiency index and the correction factor are fused to obtain a comprehensive cooperative efficiency prediction index, and the efficiency prediction threshold value is compared to output a forward-looking scheduling suggestion.

Description

Technical Field

[0001] This invention relates to the field of data mining and prediction technology, and more specifically, to a data mining and prediction method for combined cycle power units coupled with multi-source energy storage. Background Technology

[0002] In complex energy systems where gas-steam combined cycle generator units operate in conjunction with multi-source energy storage systems such as electrochemical energy storage and thermal storage tanks, the thermodynamic inertia of the unit, the charging and discharging time delays of the energy storage system, and the random fluctuations of grid dispatch commands are coupled together, forming a dynamic response characteristic with strong nonlinearity and large time delay. Existing technologies employ two main approaches: one based on thermodynamic equilibrium mechanistic models to estimate the system's operating point, but its accuracy relies on assumptions about internal parameters such as heat exchange efficiency and compressor characteristics, which are difficult to obtain in real time, making it difficult to track system state transitions under frequent multi-source energy storage operations; the other approach only extrapolates trends from historical load data in a single dimension, neglecting the constraints and support provided by boundary conditions such as the energy storage state of charge and the available enthalpy of thermal storage on the combined cycle unit's variable operating condition capabilities. Especially when dispatch commands change abruptly, existing methods lack the ability to quantitatively analyze the lag correlation between the energy storage system's response time and changes in unit energy efficiency, easily leading to significant deviations in the prediction of future adjustable output and operating efficiency, and consequently causing problems such as overcharging and over-discharging of energy storage or suboptimal load allocation of the unit. Therefore, there is an urgent need for a method that can extract dynamic correlation patterns from operational data and accurately predict the future collaborative operational efficiency of the system. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a data mining and prediction method for coupled multi-source energy storage in combined cycle units, in order to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: A data mining and prediction method for combined cycle power units coupled with multi-source energy storage includes the following steps: S1, construct the rule base R for the association between the operating characteristics of the combined cycle unit and the overall energy efficiency of the system, and the response surface model M for the state parameters of multi-source energy storage and the dynamic adjustment margin of the system; S2, collect wide-area operation data of the coupled system at T consecutive time points, extract the thermodynamic feature vectors of each sampling time point, and calculate the benchmark energy efficiency index of each time point based on the association rule base R to obtain the benchmark energy efficiency time-series vector sequence P; S3, synchronously collect the multi-source energy storage state parameters and external scheduling command parameters of the T consecutive time points to obtain the boundary condition time matrix C; S4. Based on the benchmark energy efficiency time series vector sequence P and the boundary condition time series matrix C, calculate the time lag correlation coefficient δ between the energy efficiency fluctuation sequence and the boundary parameter fluctuation sequence in each operating segment, and mark the corresponding operating segment as a period with significant lag effect when δ is greater than the preset lag judgment threshold δth. S5. Calculate the dynamic adjustment correction factor β based on the response surface model M and the boundary condition CT at the Tth time point. For the unmarked running segment, fuse the benchmark energy efficiency index PT at the Tth time point with the dynamic adjustment correction factor β to obtain the comprehensive synergistic efficiency prediction index K. Set K=0 for the marked running segment. S6 compares the comprehensive collaborative performance prediction index K with the performance prediction threshold Kth. When K is greater than or equal to Kth, it is determined that the current coupled system is in a high degree of collaborative operation and a forward-looking scheduling suggestion signal is output.

[0005] In a preferred embodiment, the construction of the rule base R for the association between the operating characteristics of the combined cycle unit and the overall energy efficiency of the system, and the response surface model M for the state parameters of multi-source energy storage and the dynamic adjustment margin of the system, specifically involves: selecting historical operating segments of the combined cycle unit under different load rate ranges and multi-source energy storage systems under different operating modes as sample clusters, and performing time-section calibration on each sample cluster; based on the time-section calibration, simultaneously collecting the gas turbine inlet temperature, steam turbine main steam flow rate, compressor inlet guide vane opening, and equivalent cycle number of electrochemical energy storage for each sample cluster. The thermal storage tank can collect data on available hot water volume, grid frequency deviation, and automatic power generation control commands, and obtain the corresponding offline calculated value of the system's comprehensive power supply efficiency. Based on the operating data, thermal characteristic quantities that can characterize the operating state of the combined cycle unit are extracted, and a rule base R for the association between unit operating characteristics and system comprehensive energy efficiency is established by combining the offline calculated value of the system's comprehensive power supply efficiency. Based on the energy storage state parameters, the change sequence of automatic power generation control commands, and the corresponding offline calculated value of the system's comprehensive power supply efficiency, a response surface model M of multi-source energy storage state parameters and system dynamic adjustment margin is established using a sub-operating condition domain fitting method.

[0006] In a preferred embodiment, the step of collecting wide-area operational data of the coupled system at T consecutive time points, extracting thermodynamic feature vectors at each sampling time point, and calculating the benchmark energy efficiency index at each time point based on the association rule base R to obtain the benchmark energy efficiency time-series vector sequence P specifically involves: synchronously collecting data from the coupled system according to a preset sampling period to obtain operational data packets of the combined cycle unit and energy storage system at each time point; performing data cleaning and timestamp alignment on the operational data packets at each time point to obtain a standardized time-series data stream; extracting compressor equivalent speed, combustion chamber equivalent temperature, and waste heat boiler node temperature difference characteristic parameters at each sampling time point based on the standardized time-series data stream, and constructing thermodynamic feature vectors at each sampling time point; inputting the thermodynamic feature vectors at each sampling time point into the association rule base R, calculating the benchmark energy efficiency index corresponding to each time point, and arranging them in chronological order to generate the benchmark energy efficiency time-series vector sequence P.

[0007] In a preferred embodiment, the step of inputting the thermal feature vectors of each sampling time point into the association rule base R and calculating the benchmark energy efficiency index corresponding to each time series point specifically involves: combining multiple thermal feature quantities corresponding to each sampling time point according to a preset entropy weight to construct a thermal feature vector for each sampling time point; performing interval normalization on the thermal feature vector and performing membership degree matching according to the preset energy efficiency cluster centers in the association rule base R to determine the membership energy efficiency domain corresponding to each sampling time point; determining the high energy efficiency cluster centers and low energy efficiency cluster centers adjacent to the membership energy efficiency domain in the association rule base R based on the membership energy efficiency domain; extracting the distance weight parameters of the thermal feature vector of the sampling time point relative to the high energy efficiency cluster centers and low energy efficiency cluster centers, and constructing an adaptive interpolation mapping function accordingly; and using the adaptive interpolation mapping function to perform continuous numerical mapping on the thermal feature vector of the sampling time point to obtain the benchmark energy efficiency index for each time series point.

[0008] In a preferred embodiment, the step of calculating the time lag correlation coefficient δ between the energy efficiency fluctuation sequence and the boundary parameter fluctuation sequence within each operating period based on the benchmark energy efficiency time series vector sequence P and the boundary condition time series matrix C specifically involves: extracting the benchmark energy efficiency index at multiple consecutive time series points from the benchmark energy efficiency time series vector sequence according to the time series, and constructing the corresponding first-order differential energy efficiency fluctuation sequence; and matching the multi-source energy storage state parameters and external scheduling command parameters of the corresponding time series points in the boundary condition time series matrix based on the time labels of the sampling time points, and constructing a boundary parameter composite fluctuation sequence corresponding to the first-order differential energy efficiency fluctuation sequence. The boundary parameter composite fluctuation sequence is shifted step by step according to a preset time delay scanning window, and the weighted mutual information value between the time-shifted boundary parameter composite fluctuation sequence and the energy efficiency first-order difference fluctuation sequence is calculated respectively; the maximum value among the weighted mutual information values ​​is selected as the time lag correlation coefficient δ of the corresponding running segment.

[0009] In a preferred embodiment, the step of calculating the weighted mutual information value between the time-shifted boundary parameter composite fluctuation sequence and the energy efficiency first-order difference fluctuation sequence specifically involves: performing detrending and amplitude standardization processing on each time-shifted boundary parameter composite fluctuation sequence to obtain a standard boundary fluctuation subsequence; extracting the effective energy efficiency fluctuation subsequence for the corresponding time period based on the energy efficiency first-order difference fluctuation sequence aligned with the standard boundary fluctuation subsequence; using the standard boundary fluctuation subsequence and the effective energy efficiency fluctuation subsequence as input, obtaining the local mutual information at each window position using a sliding entropy window calculation method, and weighting and accumulating each local mutual information according to the fluctuation intensity at the window position to obtain the weighted mutual information value corresponding to the current time-shifted order.

[0010] In a preferred embodiment, the calculation of the dynamic adjustment correction factor β based on the response surface model M and the boundary condition CT at the Tth time point specifically involves: extracting the boundary condition CT corresponding to each unlabeled running segment at the Tth time point, and inputting it into the response surface model M for interpolation response to obtain the system dynamic adjustment margin value corresponding to each unlabeled running segment; and calculating the dynamic adjustment correction factor β corresponding to each unlabeled running segment based on the system dynamic adjustment margin value according to a preset nonlinear correction rule.

[0011] In a preferred embodiment, for unmarked operating segments, the benchmark energy efficiency index PT at the T-th time point is fused with the dynamic adjustment correction factor β to obtain the comprehensive synergistic efficiency prediction index K. Specifically, the dynamic adjustment correction factor is constrained within a preset correction interval; the benchmark energy efficiency index corresponding to the T-th time point of each unmarked operating segment is fused with the corresponding dynamic adjustment correction factor to obtain the initial synergistic efficiency prediction value corresponding to each unmarked operating segment; then, based on the continuous distribution characteristics of the initial synergistic efficiency prediction value in adjacent operating domains, time-domain smoothing correction is performed to output the comprehensive synergistic efficiency prediction index K corresponding to each unmarked operating segment.

[0012] In a preferred embodiment, the step of calculating the dynamic adjustment correction factor β corresponding to each unmarked operating segment based on the system dynamic adjustment margin value according to a preset nonlinear correction rule specifically involves: inputting the system dynamic adjustment margin value corresponding to each unmarked operating segment into a preset S-shaped response generation function to obtain the corresponding initial correction factor; based on the rate of change of the automatic generation control command, the median value of the electrochemical energy storage state of charge, and the proportion of available heat in the thermal storage tank corresponding to each unmarked operating segment at the current time point, performing adjustment capability enhancement modulation on the initial correction factor to obtain an intermediate correction factor; then, based on the change gradient of the system dynamic adjustment margin value at multiple consecutive time points for each unmarked operating segment, performing time-domain smoothing constraints on the intermediate correction factor to obtain a stable correction factor; finally, performing upper and lower limit truncation and monotonic consistency correction on the stable correction factor to output the dynamic adjustment correction factor β corresponding to each unmarked operating segment.

[0013] In a preferred embodiment, the step of fusing the benchmark energy efficiency index and the corresponding dynamic adjustment correction factor corresponding to each unlabeled operating segment at the T-th time point to obtain the initial synergistic performance prediction value for each unlabeled operating segment specifically involves: extracting the benchmark energy efficiency index and the dynamic adjustment correction factor corresponding to each unlabeled operating segment at the T-th time point, and performing same-scale normalization on the benchmark energy efficiency index and the dynamic adjustment correction factor to obtain fusionable input quantities; constructing a nonlinear coupled fusion function based on the normalized benchmark energy efficiency index and the dynamic adjustment correction factor, and calculating the initial synergistic performance prediction value for each unlabeled operating segment.

[0014] The technical effects and advantages of this invention are as follows: 1. This invention constructs an association rule base R and a response surface model M, and conducts time-series lag correlation analysis by jointly using the benchmark energy efficiency time-series vector sequence P and the boundary condition time-series matrix C at T consecutive time-series points. This effectively decouples the dynamic coupling relationship between the response delay of multi-source energy storage systems and the energy efficiency changes of combined cycle units. Compared with existing methods that rely solely on thermodynamic mechanism models or single-parameter trend extrapolation, this invention introduces a time-series lag correlation coefficient δ between the energy efficiency fluctuation sequence and the boundary parameter fluctuation sequence, and marks the operating segments dominated by the energy storage charging and discharging time lag effect in combination with a preset lag judgment threshold δth. This eliminates the energy efficiency assessment distortion caused by factors such as energy storage response lag and heat transfer delay at the data mining level, and significantly reduces prediction bias.

[0015] 2. This invention introduces a system dynamic adjustment margin value based on the response surface model M, and utilizes an S-shaped response generation function, adjustment capability enhancement modulation, time-domain smoothing constraints, and nonlinear coupling fusion function to fuse the benchmark energy efficiency index PT at the T-th time series point with the dynamic adjustment correction factor β to obtain a comprehensive synergistic efficiency prediction index K. This achieves a synergistic quantitative expression of the adjustment potential of multi-source energy storage systems and the current energy efficiency level of combined cycle units. Through time-domain smoothing correction and the efficiency prediction threshold Kth determination rule, the continuity and reliability of the prediction results in the time series are further enhanced, thereby effectively improving the forward-looking prediction capability of the efficient synergistic operation state of the coupled system in future periods.

[0016] 3. This invention can not only provide accurate data decision support for load optimization allocation of combined cycle units and energy storage system charging and discharging strategies, improving the overall operating economy and frequency regulation and peak shaving performance of the system, but also reduce the extra fuel consumption and energy storage equipment lifespan loss caused by inaccurate predictions, and has significant industrial application value and low-carbon operation benefits. Attached Figure Description

[0017] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1 This is a flowchart of the data mining and prediction method for combined cycle units coupled with multi-source energy storage according to the present invention.

[0018] Figure 2 This is a flowchart of the time lag correlation coefficient calculation method of the present invention. Detailed Implementation

[0019] 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.

[0020] For examples, please refer to Figure 1 and Figure 2 As shown in this embodiment, the data mining and prediction method for coupled multi-source energy storage in a combined cycle unit includes: In this embodiment, step S1 is used to construct a rule base R for the association between the operating characteristics of the combined cycle unit and the overall energy efficiency of the system, as well as a response surface model M for the state parameters of multi-source energy storage and the dynamic adjustment margin of the system, so as to provide a unified data mining foundation for subsequent benchmark energy efficiency index calculation and dynamic adjustment correction.

[0021] Specifically, historical operating segments of combined cycle units under different load rate ranges and multi-source energy storage systems under different operating modes are first selected as sample clusters. The selection method for sample clusters is as follows: The system is divided into deep peak-shaving zones, economic operation zones, and rated output zones according to the target load rate range corresponding to the automatic generation control commands issued by the grid dispatch center. Within each load zone, multiple combinations of electrochemical energy storage in fast charging, fast discharging, and float charging / idle states, and thermal storage systems in heat storage, heat release, and idle states are selected as typical operating mode sample clusters. Each sample cluster corresponds to a continuous historical operating segment, with a duration that can be set from 15 minutes to 60 minutes. The time segment calibration is achieved through a combined time synchronization method using a wide-area phasor measurement system and a data acquisition and monitoring control system. That is, a unified time synchronization system is used to mark each operating data point with a precise time segment label to ensure that subsequent multi-source heterogeneous data can be accurately aligned along the time dimension. The offline calculation of the system's overall power supply efficiency is obtained by calculating the ratio of the total output power of the combined cycle unit to the weighted sum of the lower heating value of the fuel consumed by the gas turbine and the net output energy of the energy storage system, and by incorporating the power consumption correction term for auxiliary equipment. This yields an offline calculation value of the overall power supply efficiency that comprehensively reflects the energy efficiency level of the coupled system. This offline calculation value of the system's overall power supply efficiency serves as the benchmark label for subsequently constructing the association rule base R and the response surface model M.

[0022] After completing the time-section calibration, based on the calibration, data on the gas turbine inlet temperature, steam turbine main steam flow rate, compressor inlet guide vane opening, equivalent cycle number of electrochemical energy storage, available hot water volume in the thermal storage tank, grid frequency deviation, and automatic power generation control command data for each sample cluster are simultaneously collected, and the corresponding offline calculated value of the system's overall power supply efficiency is obtained. Specifically, the gas turbine inlet temperature, steam turbine main steam flow rate, and compressor inlet guide vane opening are collected with high precision by the combined cycle unit's distributed control system; the equivalent cycle number of electrochemical energy storage and the available hot water volume in the thermal storage tank are provided by the energy storage management system; and the grid frequency deviation and automatic power generation control command data are obtained by the synchronous phasor measurement device and the dispatch data network. For each historical operating sample cluster, continuous data collection is conducted for no less than 72 hours, with no less than 1440 time sections collected daily, thus forming a high-dimensional original training sample set containing multi-source thermal parameters, energy storage state parameters, grid boundary parameters, and offline calculated values ​​of the system's overall power supply efficiency.

[0023] After obtaining the high-dimensional original training sample set, thermodynamic feature quantities that can characterize the operating status of the combined cycle unit are extracted based on the operating data, and a rule base R for the association between the unit operating characteristics and the system comprehensive energy efficiency is established by combining the offline calculated value of the system comprehensive power supply efficiency.

[0024] In practice, wavelet threshold denoising is first applied to the gas turbine inlet temperature, steam turbine main steam flow rate, and compressor inlet guide vane opening to eliminate high-frequency white noise interference from the sensors. Then, key thermodynamic characteristics are calculated based on the combined cycle thermodynamic properties. These thermodynamic characteristics may include at least two of the following: compressor equivalent speed, combustion chamber temperature ratio, turbine entropy efficiency, waste heat boiler node temperature difference, and proximity point temperature difference. To ensure comparability between different thermodynamic characteristics, each thermodynamic characteristic is normalized according to the maximum and minimum values ​​of the overall sample. The normalization method is as follows: subtract the minimum value of the current thermodynamic characteristic from the minimum value of all samples, and then divide by the difference between the maximum and minimum values ​​of the current thermodynamic characteristic, thus ensuring that each thermodynamic characteristic falls within the 0 to 1 range.

[0025] Based on the offline calculated value of the system's comprehensive power supply efficiency, all samples are divided into three categories: inefficient, medium-efficient, and high-efficiency operating domains. Fuzzy C-means clustering algorithm is used to obtain the cluster centers, membership radii, and boundary features of each operating domain. The association rule base R is constructed as follows: the offline calculated value of the system's comprehensive power supply efficiency is used as the output axis, the normalized interval of the thermal characteristic quantity is used as the input dimension, and a continuous energy efficiency index output interval is configured for each operating domain category, where the inefficient operating domain corresponds to 0 to 0.30, the medium-efficient operating domain corresponds to 0.30 to 0.65, and the high-efficiency operating domain corresponds to 0.65 to 1. For samples located in the transition zone between adjacent operating domains, a membership-based weighted interpolation method is used to calculate the continuous energy efficiency index. That is, based on the membership weight of the current thermal characteristic vector to each cluster center, the specific position of the sample in the corresponding continuous energy efficiency index output interval is determined proportionally.

[0026] For example, if the compressor equivalent speed and combustion room temperature ratio of a certain historical operating segment fall within the transition zone between the medium-efficiency and high-efficiency operating domains, and its membership degree to the cluster center of the high-efficiency operating domain reaches 0.7, then the continuous energy efficiency index corresponding to this sample can be assigned a value of 0.71 or 0.74 to reflect that it has approached a high-efficiency operating state but has not yet completely deviated from the characteristics of medium-efficiency operation. Through the above method, an association rule base R can be formed, consisting of the normalized intervals of thermodynamic characteristics, system operating domain categories, cluster centers, and continuous energy efficiency index mapping rules.

[0027] After establishing the association rule base R, based on the energy storage state parameters, the automatic generation control command change sequence, and the offline calculated value of the corresponding system comprehensive power supply efficiency, a response surface model M of multi-source energy storage state parameters and system dynamic adjustment margin is established using a sub-operating condition domain fitting method.

[0028] The specific implementation of the sub-operating condition fitting method is as follows: First, the operating conditions are divided into a stable operating condition domain, a ramp-up operating condition domain, and a rapid load change operating condition domain according to the statistical characteristics of the automatic generation control commands. Then, within each operating condition domain, the change rate of the equivalent cycle number of electrochemical energy storage, the change rate of the available hot water volume in the thermal storage tank, the change rate of the amplitude of the automatic generation control command, and the number of command direction reversals are extracted, and combined with the offline calculated value of the corresponding system's comprehensive power supply efficiency to form a sample subset of the operating condition domain. For each sample subset of the operating domain, the change rate of the equivalent cycle number of electrochemical energy storage, the change rate of available hot water in the thermal storage tank, the change rate of the amplitude of the automatic generation control command, and the number of command direction reversals are first divided into multiple parameter grids. For example, the change rate of the equivalent cycle number can be divided into one grid at 0.1 times per hour, the change rate of available hot water can be divided into one grid at 2% per minute, and the change rate of the command amplitude can be divided into one grid at 1% of the rated load per minute. Then, the gradient of the future time period change of the offline calculated value of the system's comprehensive power supply efficiency within each parameter grid is fitted using the local weighted regression scatter smoothing method, thereby obtaining the system dynamic adjustment margin value corresponding to the parameter grid combination. The system dynamic adjustment margin value represents the potential ability of the system to improve the comprehensive energy efficiency in the future time period by adjusting the energy storage output and unit load distribution under the current energy storage state and the trend of dispatch command changes.

[0029] To improve the continuity and generalization ability of the response surface model M, multidimensional surface fitting is further performed on the system dynamic regulation margin values ​​corresponding to the parameter grid combinations under each operating domain. This multidimensional surface fitting is achieved using an adaptive kernel regression approach. Specifically, within each operating domain, the multi-source energy storage state parameters and dispatch command parameters are used as input dimensions, and the system dynamic regulation margin value is used as the output response. The optimal kernel function bandwidth parameter is determined by minimizing the cross-validation error between the fitted surface and the statistical margin value. When the rate of change of the equivalent cycle number of electrochemical energy storage is negative and its absolute value is small, the available hot water volume of the thermal storage tank is higher than 70%, and the number of automatic generation control command direction reversals is zero within a certain period, the fitted system dynamic regulation margin value is typically higher than 0.7. When the rate of change of the equivalent cycle number is large and positive, the available hot water volume is lower than 30%, and the command direction frequently reverses, the fitted system dynamic regulation margin value is typically lower than 0.25. In this way, the response surface model M can be specifically represented as a set of multidimensional mapping relationships between the operating domain category, the multi-source energy storage state parameter grid, and the system dynamic regulation margin response value.

[0030] In this embodiment, both the association rule base R and the response surface model M can be stored in the same data warehouse or in different partitions. The association rule base R includes at least five types of fields: thermodynamic feature type, normalization interval, operating domain category, cluster center, and continuous energy efficiency index mapping rule. The response surface model M includes at least six types of fields: operating domain category, equivalent cycle number change rate grid, available hot water quantity change rate grid, command amplitude change rate grid, command direction reversal number grid, and system dynamic adjustment margin response value. When subsequent steps call the association rule base R, the corresponding benchmark energy efficiency index can be directly queried or interpolated based on the input thermodynamic feature vector; when calling the response surface model M, the corresponding system dynamic adjustment margin value can be output based on the grid to which the current boundary condition belongs and the operating domain category.

[0031] For example, under the condition that the grid dispatch issues continuous ramping commands, the change rate of the equivalent cycle number of electrochemical energy storage during a certain period is -0.05 times / hour, the available hot water volume of the thermal storage tank is 85%, the change rate of the amplitude of the automatic generation control command is 0.8% per minute, and the number of command direction reversals is 0. Then, the corresponding parameter grid combination can be matched in the response surface model M, and a high system dynamic adjustment margin value can be obtained for subsequent dynamic adjustment correction factor calculation.

[0032] In this embodiment, step S2 is used to acquire wide-area operational data of the coupled system at T consecutive time points, and convert the thermal feature vectors of each sampling time point into the benchmark energy efficiency index of each time point based on the association rule base R, and further construct the benchmark energy efficiency time-series vector sequence P in chronological order. The T consecutive time points refer to T data sections set sequentially according to a fixed sampling period within the same monitoring cycle. The preset sampling period is determined comprehensively based on the thermal response time constant of the combined cycle unit and the dynamic response time of the multi-source energy storage system, preferably 5 to 60 seconds. In application scenarios with frequent grid frequency regulation requirements, the preset sampling period can be set to 10 seconds, and T equals 8640 time points can be set continuously to cover a whole day of refined operation.

[0033] When synchronously acquiring data from the coupled system according to the preset sampling period, a clock synchronization server is used for the data acquisition and monitoring control system, the energy storage management system, and the synchronous phasor measurement device to ensure that the time deviation of each data source at each time point is controlled within 50 milliseconds. After the acquisition of each time point is completed, the corresponding combined cycle unit operation data package and energy storage system operation data package are obtained respectively.

[0034] Data cleaning and timestamp alignment are performed on the runtime data packets at each time point to obtain a standardized time-series data stream. Data cleaning includes identifying and removing bad values ​​that exceed the sensor's measurement range and dead values ​​caused by communication interruptions. Outlier removal can be achieved using the median absolute deviation method, which calculates the median and its median absolute deviation for each measurement point, and marks values ​​that deviate from the median by more than five times the median absolute deviation as outliers and removes them. Missing value completion can be achieved using cubic spline interpolation of adjacent time-series points before and after the same measurement point.

[0035] Based on the standardized time-series data stream, characteristic parameters such as compressor equivalent speed, combustion chamber equivalent temperature, and waste heat boiler node temperature difference at each sampling time point are extracted, and a thermodynamic feature vector at each sampling time point is constructed. The compressor equivalent speed is obtained by calculating the ratio of the actual physical speed to the square root of the inlet total temperature; the combustion chamber equivalent temperature is obtained by calculating the difference between the combustion chamber outlet temperature and the compressor outlet temperature; and the waste heat boiler node temperature difference is obtained by calculating the difference between the evaporator outlet flue gas temperature and the steam drum saturated water temperature.

[0036] When inputting the thermal feature vectors of each sampling time point into the association rule base R and calculating the benchmark energy efficiency index corresponding to each time series point, the thermal feature quantities corresponding to each sampling time point are first combined according to the preset entropy weights to construct the thermal feature vectors of the sampling time points. The construction method of the preset entropy weights is consistent with the training samples in step S1, specifically: based on historical samples of three types of operating domains (inefficient, medium-efficient, and efficient), the information gain rate of each thermal feature quantity between different operating domains is calculated, and then normalized by dividing the information gain rate of each thermal feature quantity by the sum of the information gain rates of all thermal feature quantities to obtain the preset entropy weights corresponding to each thermal feature quantity. For example, when the information gain rates of compressor equivalent speed, combustion chamber equivalent temperature, and waste heat boiler node temperature difference account for 0.40, 0.35, and 0.25 of the total gain rate, respectively, the preset entropy weights of the three are set to 0.40, 0.35, and 0.25, respectively.

[0037] After constructing the thermal feature vector, the thermal feature vector is subjected to interval normalization, and membership matching is performed according to the preset energy efficiency cluster centers in the association rule base R to determine the energy efficiency domain corresponding to each sampling time point. During membership matching, the weighted Euclidean distance between each normalized component of the current sampling time point and each operating domain cluster center in the association rule base R is calculated one by one. The membership degree of the current sampling time point to each operating domain is calculated by dividing the reciprocal of the weighted Euclidean distance of each operating domain by the sum of the reciprocals of the weighted Euclidean distances of all operating domains. If the membership degree to a certain operating domain is greater than 0.6, then the operating domain is determined as the energy efficiency domain to which the sampling time point belongs; if the membership degree to multiple operating domains is greater than 0.6, then the cosine similarity between the current thermal feature vector and each operating domain cluster center is further compared, and the operating domain with the largest cosine similarity is determined as the energy efficiency domain to which the sampling time point belongs.

[0038] Based on the energy efficiency domain, high-energy-efficiency cluster centers and low-energy-efficiency cluster centers adjacent to the energy efficiency domain are determined in the association rule base R. After determining the high-energy-efficiency cluster centers and low-energy-efficiency cluster centers, distance weight parameters are extracted from the thermal feature vector of the sampling time point relative to the high-energy-efficiency cluster centers and low-energy-efficiency cluster centers, and an adaptive interpolation mapping function is constructed accordingly. The distance weight parameters are used to characterize the relative position of the current sampling time point between adjacent operating domains. The specific calculation logic is as follows: for the thermal feature vector, the weighted Euclidean distance between it and the low-energy-efficiency cluster center and the weighted Euclidean distance between it and the high-energy-efficiency cluster center are calculated, and the distance to the low-energy-efficiency cluster center is divided by the sum of the two distances to obtain the distance weight parameter between 0 and 1. When the distance between a sampling time point and the low-energy-efficiency cluster center is much smaller than the distance to the high-energy-efficiency cluster center, the distance weight parameter approaches 0.

[0039] When constructing the adaptive interpolation mapping function, the continuous energy efficiency index output interval corresponding to the energy efficiency domain is used as the main interval, and the distance weight parameter and the preset entropy weights of each dimension of thermodynamic features are used as function inputs. The specific implementation of the adaptive interpolation mapping function is as follows: first, the starting value and ending value of the output interval of the energy efficiency domain are linearly weighted using the distance weight parameter; then, the preset entropy weights of each dimension are fine-tuned according to the deviation between the current thermodynamic feature vector and the cluster center of the energy efficiency domain; finally, the continuous numerical mapping result of the sampling time point is generated by adding the fine-tuned weighted position coefficient to the starting value and multiplying it by the interval width. If the energy efficiency domain of a certain sampling time point at a certain time series point is the medium-efficiency operating domain, and its distance weight parameter is 0.45, and the fine-tuned weighted position coefficient is 0.40, then its continuous numerical mapping result can be calculated by adding 0.30 to 0.40 and multiplying by 0.35 to obtain 0.44. This value is the benchmark energy efficiency index of the sampling time point at that time series point.

[0040] After performing continuous numerical mapping on the thermal feature vectors at the sampling time points using the adaptive interpolation mapping function, the baseline energy efficiency indexes for each time series point are obtained and arranged in chronological order to generate a baseline energy efficiency time series vector sequence P. The baseline energy efficiency time series vector sequence P can be represented as a time series set consisting of P1, P2, up to PT, where P1 represents the baseline energy efficiency index at the first time series point, and PT represents the baseline energy efficiency index at the Tth time series point.

[0041] In this embodiment, step S3 is used to synchronously collect the multi-source energy storage state parameters and external scheduling command parameters of the T consecutive time-series points to obtain the boundary condition time-series matrix C, thereby providing a time-consistent data foundation for the subsequent calculation of the time-series lag correlation coefficient based on the benchmark energy efficiency time-series vector sequence P and the boundary condition time-series matrix C.

[0042] Specifically, during the operation of the coupled system, the equivalent cycle number of electrochemical energy storage, the available hot water volume in the thermal storage tank, and the amplitude and direction of change of automatic power generation control commands are synchronously collected through the energy storage management system and the synchronous phasor measurement device. Before generating the boundary condition time series matrix C, outlier removal and missing value completion are required for the boundary parameters collected at each time point. The boundary condition time series matrix C formed in the above manner can truly reflect the external scheduling boundary and energy storage state change process of the coupled system at T consecutive time points, and provide a unified input for subsequent time series lag correlation analysis and comprehensive synergistic efficiency prediction index calculation.

[0043] In this embodiment, step S4 is used to calculate the time lag correlation coefficient δ between the energy efficiency fluctuation sequence and the boundary parameter fluctuation sequence in each operating segment based on the benchmark energy efficiency time series vector sequence P and the boundary condition time series matrix C. When δ is greater than the preset lag judgment threshold δth, the corresponding operating segment is marked as a period with significant lag effect, so as to identify the operating segment that is significantly affected by the response delay of the multi-source energy storage system.

[0044] Specifically, the benchmark energy efficiency index at multiple consecutive time points is first extracted from the benchmark energy efficiency time series vector sequence P according to the time series. If the benchmark energy efficiency index at the first to the Tth time points in a certain period is 0.48, 0.51, 0.47, 0.53, and 0.58 respectively, then the first-order difference fluctuation sequence of energy efficiency in that period is obtained by constructing the difference between adjacent time points, that is, by subtracting the benchmark energy efficiency index of the previous time point from the benchmark energy efficiency index of the next time point, resulting in 0.03, -0.04, 0.06, and 0.05 respectively.

[0045] After obtaining the first-order difference fluctuation sequence of energy efficiency, the boundary parameter data of the corresponding time point in the boundary condition time series matrix C is matched based on the time label of the sampling time point to construct a boundary parameter composite fluctuation sequence corresponding to the first-order difference fluctuation sequence of energy efficiency. The boundary parameter composite fluctuation sequence is formed by arranging the values ​​of the first difference of the equivalent cycle number of electrochemical energy storage, the first difference of the available hot water volume of the thermal storage tank, and the first difference of the amplitude of the automatic power generation control command at multiple consecutive time points in chronological order, and then performing a linear weighted combination according to a preset weight.

[0046] In this embodiment, the boundary parameter composite fluctuation sequence is shifted step-by-step according to a preset time-delay scanning window to characterize the leading impact of energy storage system charging and discharging actions and scheduling command changes on the energy efficiency fluctuation of the combined cycle unit. The preset time-delay scanning window is determined comprehensively based on the power response time constant of electrochemical energy storage, the heat transport delay time of the thermal storage system, and the thermal inertia time of the combined cycle unit, preferably covering 1 to 20 time points. When the preset sampling period is 10 seconds, the preset time-delay scanning window corresponds to a lag range of 10 to 200 seconds. The specific implementation of the step-by-step time shift is as follows: keeping the time position of the first-order differential fluctuation sequence of energy efficiency unchanged, the boundary parameter composite fluctuation sequence is shifted forward by 1 order, 2 orders, up to 20 orders; after each time shift, the overlapping segment aligned with the time of the current time shift order result is extracted as the data interval for calculating the weighted mutual information value under the current time shift order.

[0047] For each time-shifted boundary parameter composite fluctuation sequence, detrending and amplitude standardization are performed to obtain a standard boundary fluctuation subsequence. The detrending process uses empirical mode decomposition to extract and remove the trend term from the sequence, yielding the fluctuation component sequence. The amplitude standardization is implemented using zero-mean unit variance standardization.

[0048] Based on the energy efficiency first-order difference fluctuation sequence aligned with the standard boundary fluctuation subsequence, the effective energy efficiency fluctuation subsequence for the corresponding time period is extracted. To avoid excessive amplification of local mutual information calculations due to occasional spikes in the energy efficiency first-order difference fluctuation sequence, after extracting the effective energy efficiency fluctuation subsequence, outliers exceeding three times the interquartile range of the energy efficiency first-order difference fluctuation sequence for the current time period are replaced by linear interpolation of the two nearest time points.

[0049] Using the standard boundary fluctuation subsequence and the effective energy efficiency fluctuation subsequence as inputs, the local mutual information at each window position is obtained using a sliding entropy window calculation method. The length of the sliding entropy window is determined based on the length of the current overlapping segment, preferably between 1 / 2 and 2 / 3 of the current overlapping segment length, and not less than 10 time points. At each window position, the mutual information between the standard boundary fluctuation subsequence and the effective energy efficiency fluctuation subsequence is calculated using the K-nearest neighbor entropy estimation method. Considering that different boundary parameters have different influence mechanisms on the energy efficiency fluctuation of combined cycle units, when forming the local mutual information, fluctuation sensitivity weights are assigned to the first difference of the equivalent cycle number of electrochemical energy storage, the first difference of the available hot water volume of the thermal storage tank, and the first difference of the amplitude of the automatic generation control command, respectively. The fluctuation sensitivity weights are constructed as follows: based on the original training samples used to establish the response surface model M in step S1, the Granger causal relationship significance between the fluctuation of each boundary parameter and the energy efficiency fluctuation in the future period is calculated, and then normalized by dividing the causal relationship significance of each boundary parameter by the sum of the causal relationship significance of all boundary parameters.

[0050] After obtaining the local mutual information at each window position, the local mutual information is weighted and accumulated according to the window position fluctuation intensity to obtain the weighted mutual information value corresponding to the current time shift order. The window position fluctuation intensity is the product of the variance of the standard boundary fluctuation subsequence and the variance of the effective energy efficiency fluctuation subsequence within that window position.

[0051] After performing the above calculations for each time shift order within the preset time delay scanning window, weighted mutual information values ​​corresponding to multiple time shift orders are obtained. The maximum value among these weighted mutual information values ​​is selected as the time lag correlation coefficient δ for the corresponding operating period. The time lag correlation coefficient δ characterizes the most significant time lag correlation strength between energy efficiency fluctuations and boundary parameter fluctuations during this operating period. The larger δ is, the more likely the energy efficiency changes during this period are mainly driven by the time lag effect where the energy storage system's charging and discharging actions and scheduling command changes precede the unit's energy efficiency response. A preset lag threshold δth is used to determine whether a given operating period is significantly affected by time lag. The preset lag threshold δth is constructed as follows: From the labeled samples obtained in step S1, periods where the system is in steady-state operation but the energy storage system and dispatch commands fluctuate frequently are selected as disturbance periods. Periods where the system is in a coordinated and efficient operating state are selected as coordinated periods. The distribution of the time lag correlation coefficient δ for each period in both types of samples is calculated. Then, the median value between the 80th percentile of the time lag correlation coefficient δ in the disturbance period sample and the 20th percentile of the time lag correlation coefficient δ in the coordinated period sample is taken as the preset lag threshold δth. This construction method allows the preset lag threshold δth to balance the ability to suppress time lag interference and the ability to maintain a true and efficient coordinated state.

[0052] In a set of monitoring samples of combined cycle units coupled with multi-source energy storage, the preset lag judgment threshold δth obtained by the above method can be 0.55. When the time-series lag correlation coefficient δ of a certain operating period is greater than 0.55, it indicates that the energy efficiency fluctuation and boundary parameter fluctuation of that period have a strong time-series lag correlation, and the operating period should be marked; when the time-series lag correlation coefficient δ of a certain operating period is not greater than 0.55, it is considered that the period does not belong to the case dominated by significant time-series lag effects, and can be retained for subsequent calculation of the comprehensive synergistic efficiency prediction index.

[0053] In this embodiment, step S5 is used to calculate the dynamic adjustment correction factor β based on the response surface model M and the boundary condition CT at the Tth time point, and to fuse the benchmark energy efficiency index PT at the Tth time point with the dynamic adjustment correction factor β for the unmarked operating period to obtain the comprehensive synergistic efficiency prediction index K. At the same time, K=0 is set for the marked operating period, thereby realizing the synergistic quantitative assessment of the adjustment potential of the multi-source energy storage system and the current energy efficiency level of the combined cycle unit.

[0054] Specifically, firstly, the T-th time-series boundary condition CT is extracted from the unmarked running segments in step S4. The T-th time-series boundary condition CT includes at least the rate of change of the equivalent cycle number of electrochemical energy storage, the rate of change of the available hot water volume of the thermal storage tank, the rate of change of the amplitude of the automatic generation control command, and the number of command direction reversals for the corresponding time period. Then, the boundary condition CT corresponding to each unmarked running segment is input into the response surface model M for interpolation response to obtain the system dynamic regulation margin value corresponding to each unmarked running segment. The interpolation response is implemented as follows: first, the operating domain category to which the current running segment belongs is determined, and then it is determined which set of parameter grid combinations in the response surface model M the rate of change of the equivalent cycle number of electrochemical energy storage, the rate of change of the available hot water volume of the thermal storage tank, the rate of change of the amplitude of the automatic generation control command, and the number of command direction reversals fall into; if the input boundary condition falls exactly into the parameter grid combination already stored in the response surface model M, the system dynamic regulation margin value corresponding to that parameter grid combination is directly read; if the input boundary condition is between multiple adjacent parameter grid combinations, a multidimensional linear interpolation method is used to calculate the continuous system dynamic regulation margin value.

[0055] After obtaining the system dynamic adjustment margin value corresponding to each unmarked running segment, the dynamic adjustment correction factor β corresponding to each unmarked running segment is calculated based on the system dynamic adjustment margin value according to the preset nonlinear correction rule.

[0056] The preset nonlinear correction rule first maps the system dynamic adjustment margin value to an initial correction factor using an sigmoid response generation function. The sigmoid response generation function is constructed as follows: using the system dynamic adjustment margin value as the independent variable, the initial correction factor is calculated using the formula: 0.6 + 0.6 divided by 1 + an exponential function (-10 multiplied by the difference between the system dynamic adjustment margin value and 0.5). When the system dynamic adjustment margin value is below 0.3, the initial correction factor approaches 0.65, exhibiting weak suppression; when the system dynamic adjustment margin value is above 0.7, the initial correction factor approaches 1.15, exhibiting strong enhancement; it has a high gain in the intermediate region and can sensitively reflect changes in the system adjustment margin.

[0057] After obtaining the initial correction factor, an intermediate correction factor is obtained by enhancing the regulation capability based on the rate of change of automatic generation control commands, the median value of electrochemical energy storage state of charge, and the proportion of available heat in the thermal storage tank corresponding to each unmarked operating segment at the current time point. The enhancement modulation of the regulation capability is implemented as follows: first, enhancement modulation coefficients are constructed for the rate of change of automatic generation control commands, the median value of electrochemical energy storage state of charge, and the proportion of available heat in the thermal storage tank. The enhancement modulation coefficient for the rate of change of automatic generation control commands is determined based on the unit's load change rate requirement. When the rate of change is less than 0.5% per minute, the enhancement modulation coefficient is 1.05; when the rate of change is between 0.5% and 2% per minute, the enhancement modulation coefficient is 1.00; and when the rate of change is greater than 2% per minute, the enhancement modulation coefficient is 0.92. The enhancement modulation factor for the median state of charge (SBC) of electrochemical energy storage is determined based on its ability to provide bidirectional power support. When the SBC is between 45% and 55%, the enhancement modulation factor is 1.08; when the SBC is in other ranges, the enhancement modulation factor is 0.98. The enhancement modulation factor for the usable heat ratio of the thermal storage tank is determined based on its ability to provide continuous thermal support. When the ratio is higher than 70%, the enhancement modulation factor is 1.10. Then, each enhancement modulation factor is multiplied by its corresponding weight and summed to form a comprehensive enhancement modulation factor. Finally, the initial correction factor is multiplied by this comprehensive enhancement modulation factor to obtain an intermediate correction factor.

[0058] After obtaining the intermediate correction factor, a time-domain smoothing constraint is applied to the intermediate correction factor based on the gradient of the system dynamic adjustment margin value at multiple consecutive time points for each unlabeled running segment, resulting in a stable correction factor. The gradient of the system dynamic adjustment margin value is used to characterize the rate of change of the system dynamic adjustment margin value at multiple consecutive time points during that period.

[0059] After obtaining the stable correction factor, upper and lower limit truncation and monotonic consistency correction are applied to the stable correction factor to output the dynamic adjustment correction factor β corresponding to each unlabeled running segment, and the dynamic adjustment correction factor is constrained within a preset correction interval. The preset correction interval is preferably defined as 0.70 to 1.30.

[0060] After obtaining the dynamic adjustment correction factor β corresponding to each unlabeled operating period, the benchmark energy efficiency index PT corresponding to the T-th time series point of each unlabeled operating period is fused with the corresponding dynamic adjustment correction factor β to obtain the initial synergistic efficiency prediction value corresponding to each unlabeled operating period. Specifically, the benchmark energy efficiency index PT and the dynamic adjustment correction factor β corresponding to the T-th time series point of each unlabeled operating period are first extracted, and the benchmark energy efficiency index PT and the dynamic adjustment correction factor β are normalized on the same scale to obtain the fusionable input quantity. The normalization of the dynamic adjustment correction factor β is performed linearly according to the preset correction interval, specifically: the current dynamic adjustment correction factor β is subtracted from the lower limit of the preset correction interval 0.70, and then divided by the difference between the upper limit of the preset correction interval 1.30 and the lower limit 0.70, thereby converting the dynamic adjustment correction factor β to the interval of 0 to 1.

[0061] After obtaining the fusionable inputs, a nonlinear coupled fusion function is constructed based on the normalized baseline energy efficiency index PT and the dynamic adjustment correction factor β to calculate the initial synergistic performance prediction value for each unlabeled operating segment. The nonlinear coupled fusion function comprises three parts: the first part is a linear term corresponding to the normalized baseline energy efficiency index PT; the second part is a linear term corresponding to the normalized dynamic adjustment correction factor β; and the third part is a coupling enhancement term formed by multiplying the normalized baseline energy efficiency index PT by the normalized dynamic adjustment correction factor β. The fusion weights corresponding to the three parts can be set to 0.65, 0.15, and 0.20, respectively, and the sum of the three fusion weights is 1. This design ensures that the baseline energy efficiency index PT dominates the overall evaluation, while utilizing the dynamic adjustment correction factor β to enhance or suppress the system's adjustment potential. When both are high, the coupling enhancement term significantly improves the initial synergistic performance prediction value.

[0062] After obtaining the initial collaborative performance prediction values ​​for each unlabeled runtime segment, temporal smoothing correction is performed based on the continuous distribution characteristics of the initial collaborative performance prediction values ​​in adjacent runtime domains, outputting the comprehensive collaborative performance prediction index K for each unlabeled runtime segment. The temporal smoothing correction uses a sliding median filter across 11 time points (5 before and 5 after the initial time point). For each unlabeled runtime segment, the median of the initial collaborative performance prediction values ​​across these 11 time points is used as the corrected value. The purpose of this judgment rule is to preserve the continuous distribution characteristics of the system's collaborative and efficient operating state over time, smoothing out isolated jumps caused by data noise or instantaneous disturbances. Meanwhile, runtime segments already labeled in step S4 are no longer included in the above fusion calculation and temporal smoothing correction; instead, K is directly set to 0 to explicitly exclude the interference of periods with significant temporal lag effects on the final prediction output.

[0063] In this embodiment, step S6 is used to compare the comprehensive collaborative performance prediction index K with the performance prediction threshold Kth. When K is greater than or equal to Kth, it is determined that the current coupled system is in a high degree of collaborative operation and a forward-looking scheduling suggestion signal is output.

[0064] Specifically, the comprehensive collaborative efficiency prediction index K corresponding to each time point in the coupled system's runtime domain is read point by point, and a unified judgment rule is used for threshold comparison. The efficiency prediction threshold Kth is pre-determined based on the sample data of the offline calculation value of the system's comprehensive power supply efficiency completed in step S1. The construction method is as follows: select the time period when the system is confirmed to be in a high degree of collaboration and the time period when it is in a non-high degree of collaboration, and calculate the corresponding comprehensive collaborative efficiency prediction index K distribution respectively; then, with the goal of maximizing the comprehensive score that simultaneously takes into account the high collaboration state recognition rate and the non-high collaboration state exclusion rate, the boundary value is determined as the efficiency prediction threshold Kth through the analysis of the receiver operating characteristic curve. In a set of monitoring samples of combined cycle units coupled with multi-source energy storage, the efficiency prediction threshold Kth can be set to 0.68, that is, when the comprehensive collaborative efficiency prediction index K corresponding to a certain time point is 0.68 or above, it is determined that the system is in a high degree of collaboration state at that time point; when the comprehensive collaborative efficiency prediction index K corresponding to a certain time point is less than 0.68, it is determined that the system has not yet reached a high degree of collaboration state at that time point. To avoid misjudgment caused by a single isolated high value, it can be further stipulated that: only when the comprehensive collaborative efficiency prediction index K of the target time series point reaches the efficiency prediction threshold Kth, and the comprehensive collaborative efficiency prediction index K of at least 3 consecutive time series points before and after it simultaneously reaches the efficiency prediction threshold Kth, will the target time series point be finally determined to be in a high degree of collaborative operation.

[0065] After completing the threshold comparison and determining the time series point in a high-coordination operation state, a forward-looking scheduling suggestion signal corresponding to the time series point is output. The forward-looking scheduling suggestion signal includes at least the prediction time, the comprehensive coordination efficiency prediction index K, and the recommended scheduling strategy. The prediction time uses the acquisition time of the T-th time series point, the comprehensive coordination efficiency prediction index K is the calculated K value, and the recommended scheduling strategy is generated based on the magnitude by which the comprehensive coordination efficiency prediction index K exceeds the efficiency prediction threshold Kth and the current multi-source energy storage state. Specifically, when the comprehensive collaborative efficiency prediction index K is greater than or equal to the efficiency prediction threshold Kth and less than the efficiency prediction threshold Kth plus 0.10, a Level 1 collaborative signal is output, suggesting maintaining the current automatic generation control command following strategy. When the comprehensive collaborative efficiency prediction index K is greater than or equal to the efficiency prediction threshold Kth plus 0.10 and less than the efficiency prediction threshold Kth plus 0.20, a Level 2 collaborative signal is output, suggesting moderately adjusting the charging and discharging power of the energy storage system to optimize the operating point of the combined cycle unit. When the comprehensive collaborative efficiency prediction index K is greater than or equal to the efficiency prediction threshold Kth plus 0.20, a Level 3 collaborative signal is output, suggesting proactively adjusting the load setpoint of the combined cycle unit and fully utilizing the rapid adjustment capability of the multi-source energy storage system to obtain the optimal comprehensive energy efficiency. Finally, all high-coordination operating periods can be marked on the time axis to form a forward-looking prediction map of system collaborative operation efficiency, which can be used to guide operators to make forward-looking adjustments and optimized scheduling of the system's operating strategies for future periods.

[0066] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A data mining and prediction method for combined cycle power units coupled with multi-source energy storage, characterized in that: Includes the following steps: S1, construct the rule base R for the association between the operating characteristics of the combined cycle unit and the overall energy efficiency of the system, and the response surface model M for the state parameters of multi-source energy storage and the dynamic adjustment margin of the system; S2, collect wide-area operation data of the coupled system at T consecutive time points, extract the thermodynamic feature vectors of each sampling time point, and calculate the benchmark energy efficiency index of each time point based on the association rule base R to obtain the benchmark energy efficiency time-series vector sequence P; S3, synchronously collect the multi-source energy storage state parameters and external scheduling command parameters of the T consecutive time points to obtain the boundary condition time matrix C; S4. Based on the benchmark energy efficiency time series vector sequence P and the boundary condition time series matrix C, calculate the time lag correlation coefficient δ between the energy efficiency fluctuation sequence and the boundary parameter fluctuation sequence in each operating segment, and mark the corresponding operating segment as a period with significant lag effect when δ is greater than the preset lag judgment threshold δth. S5. Calculate the dynamic adjustment correction factor β based on the response surface model M and the boundary condition CT at the Tth time point. For the unmarked running segment, fuse the benchmark energy efficiency index PT at the Tth time point with the dynamic adjustment correction factor β to obtain the comprehensive synergistic efficiency prediction index K. Set K=0 for the marked running segment. S6. Compare the comprehensive collaborative performance prediction index K with the performance prediction threshold Kth. When K is greater than or equal to Kth, determine that the current coupled system is in a high degree of collaborative operation and output a forward-looking scheduling suggestion signal. The construction of the rule base R for the association between the operating characteristics of the combined cycle unit and the overall energy efficiency of the system, and the response surface model M for the state parameters of multi-source energy storage and the dynamic adjustment margin of the system, specifically involves: selecting historical operating segments of the combined cycle unit in different load rate ranges and multi-source energy storage systems in different operating modes as sample clusters, and calibrating each sample cluster on a time cross section. Based on the aforementioned time-section calibration, the following data are simultaneously collected for each sample cluster: gas turbine inlet temperature, steam turbine main steam flow rate, compressor inlet guide vane opening, equivalent cycle number of electrochemical energy storage, available hot water volume in the thermal storage tank, grid frequency deviation, and automatic power generation control command data. The corresponding offline calculated value of the system's overall power supply efficiency is also obtained. Based on the operational data, thermodynamic characteristics characterizing the combined cycle unit's operating state are extracted. A rule base R for the association between unit operating characteristics and system overall energy efficiency is established by combining this with the offline calculated value of the system's overall power supply efficiency. Based on the energy storage state parameters, the automatic power generation control command change sequence, and the corresponding offline calculated value of the system's overall power supply efficiency, a response surface model M of multi-source energy storage state parameters and system dynamic adjustment margin is established using a sub-operating-domain fitting method.

2. The data mining and prediction method for combined cycle power units coupled with multi-source energy storage according to claim 1, characterized in that: The process involves collecting wide-area operational data of the coupled system at T consecutive time points, extracting thermodynamic feature vectors for each sampling time point, and calculating the benchmark energy efficiency index for each time point based on the association rule base R to obtain the benchmark energy efficiency time-series vector sequence P. Specifically, this involves: synchronously collecting data from the coupled system according to a preset sampling period to obtain operational data packets for the combined cycle unit and energy storage system at each time point; cleaning and aligning the operational data packets at each time point to obtain a standardized time-series data stream; extracting compressor equivalent speed, combustion chamber equivalent temperature, and waste heat boiler node temperature difference characteristic parameters at each sampling time point based on the standardized time-series data stream, and constructing thermodynamic feature vectors for each sampling time point; inputting the thermodynamic feature vectors for each sampling time point into the association rule base R, calculating the benchmark energy efficiency index corresponding to each time point, and arranging them in chronological order to generate the benchmark energy efficiency time-series vector sequence P.

3. The data mining and prediction method for combined cycle power units coupled with multi-source energy storage according to claim 2, characterized in that: The step of inputting the thermal feature vectors of each sampling time point into the association rule base R and calculating the benchmark energy efficiency index corresponding to each time series point specifically involves: combining multiple thermal feature quantities corresponding to each sampling time point according to a preset entropy weight to construct a thermal feature vector for each sampling time point; performing interval normalization on the thermal feature vector and performing membership degree matching according to the preset energy efficiency cluster centers in the association rule base R to determine the membership energy efficiency domain corresponding to each sampling time point; determining the high energy efficiency cluster centers and low energy efficiency cluster centers adjacent to the membership energy efficiency domain in the association rule base R based on the membership energy efficiency domain; extracting the distance weight parameters of the thermal feature vector of the sampling time point relative to the high energy efficiency cluster centers and low energy efficiency cluster centers, and constructing an adaptive interpolation mapping function accordingly; and using the adaptive interpolation mapping function to perform continuous numerical mapping on the thermal feature vector of the sampling time point to obtain the benchmark energy efficiency index for each time series point.

4. The data mining and prediction method for combined cycle power units coupled with multi-source energy storage according to claim 1, characterized in that: The step of calculating the time lag correlation coefficient δ between the energy efficiency fluctuation sequence and the boundary parameter fluctuation sequence in each operating period, based on the benchmark energy efficiency time series vector sequence P and the boundary condition time series matrix C, specifically involves: extracting the benchmark energy efficiency index at multiple consecutive time series points from the benchmark energy efficiency time series vector sequence according to the time series, and constructing the corresponding first-order differential energy efficiency fluctuation sequence; and matching the multi-source energy storage state parameters and external scheduling command parameters of the corresponding time series points in the boundary condition time series matrix based on the time labels of the sampling time points, and constructing a boundary parameter composite fluctuation sequence corresponding to the first-order differential energy efficiency fluctuation sequence. The boundary parameter composite wave sequence is shifted step by step according to a preset time delay scanning window, and the weighted mutual information value between the time-shifted boundary parameter composite wave sequence and the energy efficiency first-order difference wave sequence is calculated respectively. The maximum value among the weighted mutual information values ​​is selected as the temporal lag correlation coefficient δ for the corresponding runtime segment.

5. The data mining and prediction method for combined cycle units coupled with multi-source energy storage according to claim 4, characterized in that: The calculation of the weighted mutual information value between the time-shifted boundary parameter composite fluctuation sequence and the energy efficiency first-order difference fluctuation sequence specifically involves: performing detrending and amplitude standardization processing on each time-shifted boundary parameter composite fluctuation sequence to obtain a standard boundary fluctuation subsequence; extracting the effective energy efficiency fluctuation subsequence for the corresponding time period based on the energy efficiency first-order difference fluctuation sequence aligned with the standard boundary fluctuation subsequence; using the standard boundary fluctuation subsequence and the effective energy efficiency fluctuation subsequence as inputs, obtaining the local mutual information at each window position using a sliding entropy window calculation method, and weighting and accumulating each local mutual information according to the fluctuation intensity at the window position to obtain the weighted mutual information value corresponding to the current time-shifted order.

6. The data mining and prediction method for combined cycle power units coupled with multi-source energy storage according to claim 1, characterized in that: The calculation of the dynamic adjustment correction factor β based on the response surface model M and the boundary condition CT at the Tth time point is specifically as follows: extract the boundary condition CT corresponding to each unmarked running segment at the Tth time point, and input it into the response surface model M to interpolate the response, thereby obtaining the system dynamic adjustment margin value corresponding to each unmarked running segment. Based on the system's dynamic adjustment margin value, the dynamic adjustment correction factor β corresponding to each unmarked running segment is calculated according to a preset nonlinear correction rule.

7. The data mining and prediction method for combined cycle units coupled with multi-source energy storage according to claim 6, characterized in that: For the unmarked running period, the benchmark energy efficiency index PT at the Tth time point is fused with the dynamic adjustment correction factor β to obtain the comprehensive synergistic efficiency prediction index K. Specifically, the dynamic adjustment correction factor is constrained within a preset correction interval. The baseline energy efficiency index corresponding to the Tth time point of each unmarked running segment is fused with the corresponding dynamic adjustment correction factor to obtain the initial synergistic efficiency prediction value corresponding to each unmarked running segment; then, based on the continuous distribution characteristics of the initial synergistic efficiency prediction value in adjacent running domains, time-domain smoothing correction is performed to output the comprehensive synergistic efficiency prediction index K corresponding to each unmarked running segment.

8. The data mining and prediction method for combined cycle units coupled with multi-source energy storage according to claim 6, characterized in that: The step of calculating the dynamic adjustment correction factor β corresponding to each unmarked operating segment based on the system dynamic adjustment margin value according to the preset nonlinear correction rule is as follows: inputting the system dynamic adjustment margin value corresponding to each unmarked operating segment into the preset S-shaped response generation function to obtain the corresponding initial correction factor; and based on the rate of change of the automatic generation control command, the median value of the electrochemical energy storage state of charge, and the proportion of available heat in the thermal storage tank corresponding to each unmarked operating segment at the current time point, performing adjustment capability enhancement modulation on the initial correction factor to obtain the intermediate correction factor. Then, based on the gradient of the change in the system dynamic adjustment margin value at multiple consecutive time points for each unlabeled runtime segment, the intermediate correction factor is subjected to time-domain smoothing constraint to obtain the stable correction factor; finally, the stable correction factor is subjected to upper and lower limit truncation and monotonic consistency correction to output the dynamic adjustment correction factor β corresponding to each unlabeled runtime segment.

9. The data mining and prediction method for combined cycle power units coupled with multi-source energy storage according to claim 7, characterized in that: The step of fusing the benchmark energy efficiency index and the corresponding dynamic adjustment correction factor for each unlabeled operating segment at the T-th time point to obtain the initial synergistic performance prediction value for each unlabeled operating segment is as follows: extracting the benchmark energy efficiency index and the dynamic adjustment correction factor for each unlabeled operating segment at the T-th time point, and performing same-scale normalization on the benchmark energy efficiency index and the dynamic adjustment correction factor to obtain fusionable input quantities; constructing a nonlinear coupled fusion function based on the normalized benchmark energy efficiency index and the dynamic adjustment correction factor to calculate the initial synergistic performance prediction value for each unlabeled operating segment.

Images