A power generation dynamic prediction method and system fusing weather and equipment state

Abstract

The application relates to the technical field of power generation prediction, and discloses a power generation dynamic prediction method and system fusing meteorological and equipment state. The method synchronously collects meteorological environment and equipment state data, extracts meteorological features and equipment features, constructs cross features, and fuses into decoupling feature vectors through principal component analysis; the fusion feature sequence is input into a gated recurrent unit network, a preliminary prediction value is output, and model parameters are continuously updated through online gradient descent; dual physical benchmarks of instant and inertia are constructed, physical deviation degrees are quantified, a dynamic correlation weight matrix is used to construct a correction confidence, and transient and inertia correction amounts are generated in combination with equipment health factors to link and correct the preliminary prediction value, and finally output the final prediction value. The application realizes data deep fusion and model self-adaptation, and improves the prediction accuracy and robustness under complex working conditions.

Description

Technical Field

[0001] This application relates to the field of power generation prediction technology, specifically a method and system for dynamic prediction of power generation that integrates meteorological and equipment status. Background Technology

[0002] Traditional power generation forecasting methods often rely solely on meteorological data, resulting in a limited forecast dimension. They fail to adequately consider the operational status of the power generation equipment, neglecting the impact of varying operating conditions such as equipment aging, component wear, and malfunctions on power generation efficiency, leading to significant forecast deviations. Furthermore, traditional forecasting models have fixed parameters and lack dynamic adaptive adjustment capabilities, making them unsuitable for scenarios involving fluctuating equipment conditions and complex weather changes, resulting in poor generalization ability. In addition, traditional methods lack a two-dimensional data fusion analysis mechanism, failing to promptly link meteorological changes with abnormal equipment conditions to correct forecast values. This hinders high-precision dynamic forecasting and fails to meet the practical needs of refined power plant scheduling and operation and maintenance management. Summary of the Invention

[0003] (a) Technical problems to be solved To address the shortcomings of existing technologies, this application provides a method, system, medium, and processor for dynamic power generation prediction that integrates meteorological and equipment status data. This method overcomes the limitations of traditional methods, such as single data, fixed models, and lack of linkage, and achieves high-precision, robust, and physically consistent dynamic power generation prediction. It solves the problems of the inability to promptly link and correct predicted values ​​when meteorological changes and equipment operating abnormalities occur, the difficulty in achieving high-precision dynamic prediction, and the inability to meet the actual needs of refined scheduling and operation and maintenance management of power plants.

[0004] (II) Technical Solution To achieve the above objectives, this application provides the following technical solution: Firstly, this application provides a method for dynamic prediction of power generation that integrates meteorological and equipment status, including the following steps: Meteorological and environmental data and power generation equipment status data are collected simultaneously and combined into a dual-source synchronous dataset after preprocessing. Meteorological impact features and power generation equipment status features are extracted from the dual-source synchronous dataset. Deep fusion and correlation modeling of the dual-source data are performed to obtain the fused feature vector and generate a dynamic correlation weight matrix. Using the fused feature vector as input, a preliminary predicted value is obtained by dynamically predicting power generation through a dynamic adaptive prediction model. A dual reference system of instantaneous physical reference and inertial physical reference is constructed to quantify the degree of deviation from the actual physics. The reliability of the deviation is evaluated using a dynamic correlation weight matrix. Based on the dual reference system and the reliability, an adaptive correction amount is generated to perform a linkage correction on the preliminary prediction value to obtain the final prediction value.

[0005] Furthermore, the step of extracting meteorological impact features and power generation equipment status features from the dual-source synchronous dataset, and performing deep fusion and correlation modeling of the dual-source data to obtain a fused feature vector, specifically includes the following steps: Extract meteorological impact feature vectors and power generation equipment status feature vectors from the dual-source synchronous dataset; Normalized cross features are obtained by cross-constructing the meteorological impact feature vector and the power generation equipment status feature vector; The original meteorological impact feature vector and the power generation equipment status feature vector are concatenated with the cross features to form the original high-dimensional feature vector; The original meteorological impact feature vector, power generation equipment status feature vector, and original high-dimensional feature vector are dynamically constructed into a covariance matrix based on their respective historical sliding windows, and principal component analysis is performed. The number of principal components is adaptively selected based on the cumulative contribution rate threshold for independent dimensionality reduction. The dimensionality reduction results of each group are then concatenated into the final fused feature vector to avoid multicollinearity among meteorological features, power generation equipment status features, and cross features, which could lead to instability in the training of the GRU model.

[0006] Furthermore, the meteorological impact characteristics include effective irradiance, ambient temperature, and wind speed fluctuation rate; the power generation equipment status characteristics include component degradation rate, bearing temperature rise rate, and inverter efficiency deviation.

[0007] Furthermore, the generation of the dynamic correlation weight matrix refers to constructing a dynamic correlation weight matrix by calculating the sliding Pearson correlation coefficient between each component of meteorological impact characteristics and each component of power generation equipment status characteristics over a certain period of time, which serves as a statistical baseline for subsequent judgment on whether the meteorological-equipment linkage is abnormal.

[0008] Furthermore, the step of using the fused feature vector as input to obtain a preliminary predicted value for power generation through a dynamic adaptive prediction model includes the following steps: The fused feature vector at the current moment and the sequence of fused feature vectors from the previous several historical moments constitute the input sequence tensor; The input sequence tensor is expanded by time step and input into the prediction model to obtain preliminary prediction values ​​for power generation.

[0009] Furthermore, the construction of a dual reference system of instantaneous physical reference and inertial physical reference to quantify the degree of deviation from the actual physics, the evaluation of the reliability of the deviation using a dynamic correlation weight matrix, and the generation of an adaptive correction amount based on the dual reference system and the reliability to perform a linkage correction on the preliminary prediction value to obtain the final prediction value, includes the following steps: Real-time construction of instantaneous physical reference power and inertial physical reference power as dual-scale physical reference benchmarks; Determine the intensity of physical field mutations and equipment health factors based on instantaneous physical reference power and inertial physical reference power; The corrected confidence level is determined based on the dynamic correlation weight matrix; The transient correction and inertial correction are calculated based on the instantaneous physical reference power, inertial physical reference power, physical field abrupt change intensity, equipment health factor, and correction confidence level. The initial prediction value is corrected based on the transient correction and the inertial correction to obtain the final prediction value.

[0010] Furthermore, the determination of the corrected confidence level based on the dynamic correlation weight matrix includes the following steps: The coupling strength between meteorological conditions and power generation equipment operating conditions is determined based on the dynamic correlation weight matrix; Then, the temporal stability of the meteorological-power generation equipment linkage relationship is obtained based on the dynamic correlation weight matrix; The corrected confidence level is obtained based on the coupling strength and timing stability.

[0011] Secondly, this application provides a dynamic power generation prediction system that integrates meteorological and equipment status data, used to implement the above-described dynamic power generation prediction method that integrates meteorological and equipment status data, including: The data acquisition module is used to simultaneously acquire meteorological and environmental data and power generation equipment status data, which are then combined into a dual-source synchronous dataset after preprocessing. The fusion module is used to extract meteorological impact features and power generation equipment status features from the dual-source synchronous dataset, perform deep fusion and correlation modeling of the dual-source data to obtain the fused feature vector, and generate a dynamic correlation weight matrix. The prediction module is used to obtain preliminary prediction values ​​by dynamically predicting power generation through a dynamic adaptive prediction model, taking the fused feature vector as input. The correction module is used to construct a dual reference system of instantaneous physical reference and inertial physical reference to quantify the degree of deviation from the actual physics, and to evaluate the credibility of the deviation using a dynamic correlation weight matrix. Based on the dual reference system and the credibility, an adaptive correction amount is generated to perform linkage correction on the preliminary prediction value to obtain the final prediction value.

[0012] (III) Beneficial Effects Compared with existing technologies, this application provides a method, system, medium, and processor for dynamic prediction of power generation that integrates meteorological and equipment status data, which has the following beneficial effects: 1. Overcoming the limitations of a single data source: Simultaneously collect meteorological and equipment status data, extract multidimensional features from both sources and construct them cross-referenced. Through PCA orthogonal transformation, multicollinearity is eliminated, achieving deep fusion of dual sources and feature decoupling. This enables the model to clearly distinguish whether the power generation deviation originates from meteorology or equipment, avoiding mutual interference between compensation strategies.

[0013] 2. Construct a dynamic adaptive prediction model: A GRU network is adopted and an online gradient descent mechanism is introduced. After each prediction, the actual value is read back to update the network weights in real time, so that the model can continuously self-optimize as the equipment ages and the weather changes, ensuring the prediction accuracy and reliability under various working conditions.

[0014] 3. Pioneering Dual-Physical Reference Correction Mechanism: A dual physical reference power mechanism is constructed, integrating instantaneous and inertial power. By quantifying the deviation between the two and generating a correction quantity carrying direction and amplitude information through hyperbolic tangent mapping, the prediction results achieve both high accuracy and clear physical interpretability. This prevents purely data-driven models from lacking physical prior constraints and thus avoiding non-causal prediction biases that may exceed the upper limit of equipment physical capacity or mismatch with irradiation energy input under extreme weather conditions.

[0015] 4. Introduce a dynamic correlation weight matrix as a confidence gate: Construct a dynamic correlation matrix based on the sliding Pearson correlation coefficient, extract the coupling strength and time series stability and multiply them to obtain the modified confidence. The modification is released only when both conditions are met, realizing the autonomous on / off of the modification excitation, achieving dual safety redundancy, and preventing sensor or communication anomalies from contaminating the prediction results.

[0016] 5. Achieve dual-timescale collaborative prediction: Simultaneously output prediction values ​​for the next 15 minutes and 60 minutes, with transient rapid follow-up correction and inertial inheritance decay correction designed respectively, coupled with an independent aging penalty term, so that it can respond sensitively to sudden changes in the short term and reflect trends steadily in the medium and long term, fully meeting the scheduling and operation and maintenance needs.

[0017] 6. Fully continuous threshold-free smooth design: All intermediate variables and final corrections use continuous functions (saturation mapping, exponential decay, etc.) to avoid prediction jumps and oscillations caused by hard cutoff thresholds at critical points, thereby improving the reliability of the system in actual engineering. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the process structure of a dynamic power generation prediction method that integrates meteorological and equipment status data according to this application.

[0019] Figure 2 This is a schematic diagram of the process structure for obtaining the fused feature vector through deep fusion and correlation modeling of dual-source data for this application.

[0020] Figure 3 This is a schematic diagram of the process structure for the linkage correction of the preliminary predicted values ​​in this application.

[0021] Figure 4 This is a schematic diagram of the structure of a dynamic power generation prediction system that integrates meteorological and equipment status data according to this application. Detailed Implementation

[0022] 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.

[0023] In the description of this invention, it should be understood that the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0024] In addition, a fixed connection refers to a connection in which parts or components are fixed and there is no relative movement; a transmission connection refers to a connection in which mechanical motion or torque is transmitted to other working parts through a transmission component; a sliding connection refers to a connection in which two objects are in contact but not fixed and can slide relative to each other; and a rotational connection refers to a connection in which two objects are in contact but not fixed and can rotate relative to each other.

[0025] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0026] Example 1 like Figures 1 to 3 As shown, this embodiment provides a method for dynamic prediction of power generation that integrates meteorological and equipment status data, including the following steps: S1: Simultaneously collect meteorological and environmental data and power generation equipment operation status data, and combine them into a dual-source synchronous dataset after preprocessing.

[0027] S11: Specifically, the output terminals of the meteorological sensor group deployed at the power plant (including irradiance meters, anemometers, and thermo-hygrometers), the data registers of the SCADA system on the power generation equipment side, and the vibration / temperature monitoring terminals are connected to different input channels of the data acquisition unit. Sampling is triggered by the same clock source to ensure strict synchronization of the start times of all data input channels. Since the sampling frequencies of each channel are different, resulting in data streams with different frequencies, subsequent interpolation / alignment operations are required to resolve frequency mismatch issues. Let the meteorological data sampling frequency be... The sampling frequency of device status data is To ensure the time alignment accuracy of data from both sources (two data acquisition sources), the following settings are configured: and The ratio is an integer, and linear interpolation is performed on the lower sampling frequency based on the higher sampling frequency. For example, when , At that time, using 10Hz as the reference, the meteorological data was interpolated to 10Hz using three splines to ensure that the two data sources have a unified time resolution.

[0028] S12: After sampling, a moving average filter is used to denoise the original sampled data. The filter window width is set to 5 sampling points. The filtered data forms a time-aligned meteorological dataset. and device status dataset ,in Incrementing in seconds (or other units), They represent irradiance respectively. Ambient temperature Wind speed , These represent the output power. Vibration amplitude Temperature rise rate Inverter efficiency .Will and Combined into a dual-source synchronized dataset .

[0029] S2: Extract meteorological impact features and power generation equipment status features from the dual-source synchronous dataset, and perform deep fusion and correlation modeling of the dual-source data to obtain the fused feature vector. And generate a dynamic correlation weight matrix. .

[0030] Specifically, such as Figure 2 As shown, the dual-source synchronization dataset obtained from S1 Extract meteorological impact feature vectors from them. and power generation equipment state feature vector In this step, The data sampling time interval (in this embodiment, it is taken as...) ), To integrate the feature vector update period (in this embodiment, we take...) ).

[0031] (a) From dual-source synchronized datasets Extracting meteorological impact feature vectors .

[0032] Meteorological Influence Feature Vector It consists of three components: effective irradiance Ambient temperature and wind speed fluctuation rate ,Right now: ; The specific extraction methods for the three components are as follows: (1) Effective irradiance Because the power response of power generation equipment has thermal and electrical inertia, fluctuations in instantaneous irradiance will not be immediately and fully reflected in the output power. Therefore, the old method is used... The average value of each sampling point is used as the effective irradiance characteristic value: ; In the formula, For the past The instantaneous irradiance values ​​at each sampling time, where For sampling point index ( ), This represents the number of sampling points included in the past 10 minutes. For example, when... hour, .

[0033] (2) Ambient temperature The ambient temperature data obtained from S1 after synchronous acquisition and moving average filtering is used directly. As a feature component in this step, no additional processing is required.

[0034] (3) Wind speed fluctuation rate Defined as past The ratio of the standard deviation of wind speed to the average wind speed within each sampling point: ; In the formula, For the past The instantaneous wind speed values ​​at each sampling time, where For sampling point index ( ), For the past The average wind speed at each sampling point This represents the number of sampling points included in the past minute. It is a small positive number (with the same dimensions as wind speed) to prevent the denominator from being zero.

[0035] (ii) From dual-source synchronized datasets Extracting the state feature vector of power generation equipment .

[0036] Power generation equipment state feature vector It consists of three components: component attenuation rate Bearing temperature rise rate Inverter efficiency deviation ,Right now: ; The specific extraction methods for the three components are as follows: (1) Component attenuation rate Defined as the relative deviation between the actual output power and the theoretical output power at the current moment: ; In the formula, This represents the actual output power at the current moment (read by the SCADA system in S1). To be based on the current irradiance The calculated theoretical output power (not a measured value, but an intermediate derived value calculated within this step) is calculated as follows: ; In the formula, The effective light-illuminated area of ​​the component. The component conversion efficiency under standard test conditions. The temperature power coefficient of the component. This is the standard test temperature.

[0037] (2) Bearing temperature rise rate Defined as past The slope of the bearing temperature change within each sampling point is obtained by least-squares linear fitting: ; In the formula, For the past The bearing temperature sample values ​​at each sampling time, where For sampling point index ( ), This represents the number of sampling points included in the past 2 minutes. The unit is .

[0038] (3) Inverter efficiency deviation Defined as the deviation between the inverter's current actual conversion efficiency and its rated efficiency. ; In the formula, This refers to the nominal conversion efficiency of the inverter under rated operating conditions. This represents the current AC output power of the inverter. This represents the current DC-side input power of the inverter, both of which are read by the SCADA system in S1.

[0039] (III) Meteorological impact feature vector With the state feature vector of power generation equipment Normalized cross features are obtained by constructing cross features. .set up express The 1 component (of which (corresponding to effective irradiance, ambient temperature, and wind speed fluctuation rate, respectively). express The 1 component (of which (corresponding to component degradation rate, bearing temperature rise rate, and inverter efficiency deviation respectively), then the... The meteorological characteristics and the first Cross-characteristics of the state features of individual power generation equipment for: ; In the formula, and These are the maximum values ​​of their respective components over the past 24 hours. This is for normalized cross features. Cross feature vector. From all Composed of several cross-feature components: ; (iv) Generation of fused feature vectors Meteorological feature vectors Device status feature vector and cross feature vectors Principal component analysis was performed to reduce dimensionality.

[0040] by For example, the sample matrix is ​​constructed using the 96 historical values ​​preceding the current moment (corresponding to the past 24 hours). Calculate the covariance matrix ,right Perform eigenvalue decomposition ,in It is an eigenvalue diagonal matrix and , This is the corresponding eigenvector matrix; It is the covariance matrix The three eigenvectors are arranged according to their corresponding eigenvalues. Arrange from largest to smallest. Select those that satisfy... Minimum number of principal components ,Pick The former Columns form the projection matrix Then the features after dimensionality reduction are: ; and Similarly, we obtain the following results: and The three elements are concatenated to form the final fused feature vector: ; like , , ,but It is 9-dimensional.

[0041] Grouped PCA eliminates multicollinearity within each group and reduces the input dimension from 15 to 9, thus lowering model complexity. It not only eliminates multicollinearity within groups to ensure stable model training, but also deliberately preserves the original correlation between meteorological and equipment groups, providing a true statistical baseline for the subsequent dynamic association weight matrix. Simultaneously, the grouping strategy avoids the problem of equipment degradation signals being overwhelmed by meteorological big data due to significant differences in scale during global dimensionality reduction, and the erosion of the original feature space by cross-features, ensuring clear and traceable physical origins. Furthermore, dynamic covariance updates based on a sliding window adaptively adjust the dimensionality reduction benchmark as operating conditions shift, improving the timeliness and representational capability of features in complex field environments.

[0042] (v) At the same time, by calculating the sliding Pearson correlation coefficients between each component of meteorological impact characteristics and each component of power generation equipment status characteristics over a certain period of time, a dynamic correlation weight matrix is ​​constructed. This serves as a statistical baseline for subsequent assessments of whether the meteorological-equipment linkage is abnormal. Specifically, it involves calculating meteorological impact characteristics. Characteristics of power generation equipment Dynamic correlation weight matrix between , its first Line 1 Column elements (in correspond The three components, correspond The three components, the meaning of the subscripts, and the construction of feature intersections. , (Completely identical) Calculated by the following formula: ; In the formula, Indexing historical moments ( ), For the feature vector update period, This is for the number of statistical periods (corresponding to the past 24 hours). and Each in the past The mean of the corresponding component within each time point. (From all) element The complete matrix formed is: ; The matrix It reflects the typical linkage between meteorological factors and equipment response under normal operating conditions, and serves as the statistical baseline for S4 to determine whether the linkage between the two is abnormal at the current moment.

[0043] Will The S3 is fed in real time as input to the dynamic power generation prediction model. The data is fed into S4 in real time as a baseline reference matrix for abnormal linkage determination.

[0044] S3: Using fused feature vectors As input, a preliminary forecast value is obtained by dynamically predicting power generation using a dynamic adaptive GRU prediction model. .

[0045] Specifically, this step is the core execution step for dynamic prediction of power generation. Based on the fused feature vector obtained in S2... At the current moment of and its predecessor A sequence of historical moments The input sequence tensors that make up the prediction model ,in The update period for the feature vector is . This is the length of the sliding window (covering the past 12 hours). The input sequence tensor... The data is fed into a three-layer gated recurrent unit (GRU) network for power generation prediction, as detailed below: Input sequence tensor Expand by time step, let the first... Input at each time step (Right now The (line), then the GRU network in the 1st row The update process for each time step is as follows: ; ; ; ; In the formula, To update the gate output vector, To reset the gate output vector, Let be the candidate hidden state vector. To hide the state vector at the current moment, The hidden state vector from the previous time step. For the weight matrix ( (for hidden layer dimensions) For bias terms, It is the sigmoid activation function. The hyperbolic tangent activation function is used. For element-wise multiplication, It is a vector consisting entirely of 1s.

[0046] The weight matrix of the GRU network and Random initialization is performed using a Xavier uniform distribution, with bias terms... Initialize to an all-zero vector. Initial hidden state. Set as the zero vector. Before network deployment, offline pre-training is required using historical power plant data. After pre-training, all parameters are loaded into the online prediction system as the initial state, and then continuously updated using the online gradient descent method described in this step.

[0047] After the network propagates forward, the output layer generates the future... and Preliminary power generation forecasts at two time scales: ; In the formula, This is the initial predicted value vector. and These are the predicted power generation values ​​for the next 15 minutes and 1 hour, respectively. To output the weight matrix, This is the output bias term.

[0048] After the forecast is completed, the actual power generation at the forecast time is read back from the power station's electricity meters. The prediction bias is calculated, and the GRU network weights are dynamically updated using online gradient descent. ; ; In the formula, As the baseline prediction deviation vector, , and These are the sets of network weights before and after the update. For learning rate, This is used to predict the gradient of the sum of squared biases (i.e., the loss function) with respect to all weights. This online update mechanism enables the prediction model to continuously and adaptively adjust itself as the device ages.

[0049] S4: Construct a dual reference frame consisting of an instantaneous physical reference and an inertial physical reference to quantify the degree of deviation from the actual physical reality. Utilize a dynamic correlation weight matrix to assess the reliability of the deviation. Based on the dual reference frame and the reliability assessment, generate an adaptive correction to the initial predicted value. Perform linkage corrections to obtain the final predicted value. Output to EMS.

[0050] like Figure 3 As shown, this step quantifies the true physical deviation of the current weather field from its historical inertial state by constructing a dual reference system of instantaneous and inertial physical references. The reliability of this deviation is then evaluated using the dynamic correlation weight matrix output by S2, ultimately generating a continuous and smooth adaptive correction. All intermediate variables and the final correction are continuous functions, ensuring that the predicted value has no jumps or oscillations. The final output is the predicted value. .

[0051] S41: Real-time construction of instantaneous physical reference power and inertial physical reference power as a dual-scale physical reference benchmark, as detailed below: The power output of a photovoltaic (PV) power generation system is simultaneously constrained by two physical constraints: the immediate potential output determined by the current instantaneous irradiance and temperature, and the actual achievable response determined by the thermal and electrical inertia of the modules. The former changes rapidly, reflecting the immediate fluctuations of the meteorological field; the latter changes slowly, representing the stable output baseline of the system under the current thermal conditions. Therefore, this step constructs physical reference benchmarks for two time scales.

[0052] (1) Instantaneous physical reference power : ; This step utilizes the current instantaneous irradiance at the second level. and instantaneous temperature Calculate the potential maximum output at the current moment. In the formula... The effective light-illuminated area of ​​the component. The component conversion efficiency under standard test conditions. The temperature power coefficient of the component. For the standard test temperature (25°C), this formula specifically incorporates the currently measured inverter efficiency. As a correction term, it is defined as follows: ; in and These are the current AC side output power and DC side input power of the inverter, which are read in real time by the SCADA system in S1. The rated efficiency of the inverter; It is a very small positive number (which can be taken as) This is used to prevent the denominator from being zero. Taking the minimum value is to prevent measurement anomalies from causing artificially high efficiency. The characteristic of this benchmark is its extremely fast response, which can sensitively capture minute-level meteorological changes such as cloud formations passing by and sudden rain showers.

[0053] (2) Inertial physical reference power : ; This formula uses the moving average of irradiance over the past 10 minutes. Temperature sliding average and inverter efficiency moving average Substituting into the same physical equation, the 10-minute timescale is comparable to the thermal time constant of the photovoltaic module, reflecting the stable power output level of the system under the current thermal equilibrium state, which will not fluctuate drastically due to a few seconds of cloud shadow passing by. This benchmark represents the system's expected inertial output and serves as a reference point for judging whether the current moment represents a true physical abrupt change.

[0054] The collaborative logic of dual benchmarks: when and When the values ​​are close, it indicates that the current meteorological conditions are consistent with the average state over the past 10 minutes, the system is operating smoothly, and no significant correction is needed. When the two values ​​deviate significantly, it indicates that a real and continuous change has occurred in the meteorological field (distinct from single-point noise), and correction should be initiated. The difference between the two values ​​includes both the direction of change (positive for enhancement, negative for weakening) and the magnitude of change, providing a physical basis for subsequent corrections.

[0055] S42: Determine the intensity of physical field mutations and equipment health factors based on instantaneous physical reference power and inertial physical reference power, as detailed below: (1) Physical field abrupt change intensity : ; This step uses the relative deviation between the instantaneous reference and the inertial reference as the independent variable, and applies it through the hyperbolic tangent function. Mapped to Interval. The function has two excellent characteristics: firstly, it is an odd function with symmetry, where positive values ​​indicate an enhancement of the physical field (such as an increase in irradiance) and negative values ​​indicate a weakening of the physical field (such as cloud cover), and the sign naturally carries information about the correction direction; secondly, it is saturated nonlinear, where the output tends to saturate (close to ±1) when the deviation exceeds a certain range, avoiding excessive correction under extreme weather conditions. This is consistent with actual physical laws - photovoltaic output power has a physical upper limit (limited by the rated power of the module) and cannot be infinitely proportional to the increase in irradiance.

[0056] molecular The absolute difference between the current instantaneous physical reference and the inertial physical reference of the past 10 minutes was directly calculated. The difference between the two is the original deviation of the meteorological field relative to its own historical inertial state, and the unit is kW / MW.

[0057] In the denominator Relative deviation normalization was achieved, making Regardless of the installed capacity of the power plant, the same set of parameters can be applied to power plants of different capacities.

[0058] (2) Equipment health degradation factors : ; in, It is the cumulative weighted average efficiency of the "effective power generation period" over the past 24 hours up to the current moment, and its recursive update formula is: ; And an integral adaptive learning rate is introduced. : ; in, The reference power generation per hour at rated installed capacity (e.g.) ), used for normalization. For a saturation function, the lower limit is... A value of 0.6 can be used (to prevent excessive drop in the factor due to extreme weather on a single day), with an upper limit of 1. A time step of 15 minutes can be used, consistent with the original system's time step.

[0059] The working principle is as follows: In the subsequent correction formula, As a multiplicative factor in the transient correction term, the healthier the equipment, the more adequate the correction response; the more severe the equipment degradation, the more the correction intensity is automatically reduced, preventing aging equipment from being overcorrected to an output level that it cannot actually reach during sudden irradiation changes, reflecting the principle of correction based on capability.

[0060] S43: Determine the corrected confidence level based on the dynamic correlation weight matrix.

[0061] S2 output dynamic correlation weight matrix This reflects the typical linkage between meteorological factors and equipment response under normal operating conditions. This step extracts two continuous indicators from this data to construct a modified confidence level. It is used to determine whether the current physical benchmark is trustworthy.

[0062] (1) Determine the coupling strength between meteorological conditions and power generation equipment operating conditions based on the dynamic correlation weight matrix. : ; This indicator is The average of the sum of the absolute values ​​of all elements reflects the overall linear correlation strength between meteorological characteristics and equipment condition characteristics. In a normally operating power plant, irradiance is highly correlated with output power, therefore... Usually close to 1; when A significantly lower reading indicates a disruption in the weather-equipment coupling relationship. Possible causes include: the irradiator being blocked or malfunctioning, SCADA communication interruption, or the equipment operating under nonlinear extreme conditions. In this case, the reliability of the physical reference decreases, and corrections should be reduced.

[0063] (2) Then, the temporal stability of the meteorological-power generation equipment linkage relationship is obtained based on the dynamic correlation weight matrix. : ; in The update period is consistent with the feature vector update period in S2; This is the Frobenius norm (the square root of the sum of the squares of all elements in the matrix). This metric measures the drastic change in the correlation matrix between two adjacent update cycles: if... If the change is minimal compared to 15 minutes ago, then... A value close to 1 indicates a stable meteorological-equipment linkage and reliable physical reference; if... If a drastic change occurs (such as a sudden drop in the irradiator's power without a synchronous decrease), then If the value rapidly approaches 0, the correction intensity should be significantly reduced, or even the correction should be paused, to avoid misinterpreting sensor malfunctions as real physical changes.

[0064] Reason for choosing exponential form: Exponential functions guarantee It is also sensitive to changes in input. As the rate of change of the norm gradually increases from 0, The smooth descent, rather than a hard cutoff, aligns with the design principle of a fully continuous, threshold-free system.

[0065] (3) Obtain the modified confidence level based on coupling strength and time series stability. : ; Multiplying the two together yields The corrected confidence level for the interval. Only when both conditions of high coupling strength and time stability are simultaneously met... When the value approaches 1, the correction amount is fully released; if any condition worsens, Automatic contraction suppresses the correction force. This design creates dual safety redundancy: low coupling strength (sensor failure) or matrix mutation (abnormal linkage) can independently trigger weight reduction, ensuring that erroneous corrections do not contaminate the prediction results.

[0066] S44: Calculate the transient correction and inertial correction based on the instantaneous physical reference power, inertial physical reference power, physical field mutation intensity, equipment health factor, and correction confidence level.

[0067] (1) Transient correction (Prediction for the next 15 minutes): ; in In the original text S2, it is defined as: ; In addition, the rightmost factor This is an amplitude adjustment factor, with a value range of... Its design philosophy is: when the deviation between the two benchmarks is very small (close to 0), the exponential part... When the factor approaches 0, no correction is triggered; as the deviation gradually increases, the factor monotonically increases and saturates towards 1, with a smooth saturation curve and no hard cutoff. In the denominator... Adaptive scaling under strong light conditions was achieved. Larger absolute deviations are needed to achieve the same relative correction weights under low light conditions. If the cloud cover is small, the threshold will automatically decrease, which is consistent with the actual physical characteristics of photovoltaic systems: 100W / m² of cloud cover at noon. 2 The power impact caused by absolute changes is far greater than that of the same changes during the early morning hours; therefore, the correction weights should be adapted to the current illuminance level.

[0068] Multiply : The sign of the correction determines the direction of the correction—positive (enhancement) or negative (reduction), and its absolute value determines the directional strength of the correction. Already by Saturation The range ensures that the direction is clear and the amplitude is controllable.

[0069] Multiply : Convert the dimensionless correction weight into a correction amount with power dimensions (kW / MW). This represents the theoretical power output level under current lighting conditions. Using this as the amplitude benchmark ensures that the absolute size of the correction amount naturally matches the current power generation level of the power station: the correction amplitude is large when the light is strong, small when the light is weak, and automatically resets to zero at night.

[0070] Multiply The equipment health factor acts as a multiplicative coefficient; the better the equipment's health status (…), the greater the benefit. The closer to 1), the more thorough the correction; the more severe the equipment degradation, the more the correction intensity is automatically reduced. This reflects the physical reality that even with the best irradiation, aging equipment will not output as much as healthy equipment.

[0071] Multiply The confidence level serves as the final gate coefficient, directly shutting down or weakening the correction channel in the event of sensor failure or linkage anomaly.

[0072] (2) Inertia correction amount (Prediction for the next 60 minutes): ; in (Time interval between 15-minute and 60-minute predictions) This is the decay time constant.

[0073] The logical structure of this expression consists of two parts: First item (transient inheritance): Medium- to long-term forecasts cannot completely ignore current physical mutations, but they should not be as sensitive as short-term forecasts. The exponential decay factor (approximately...) The transient correction was passed on to the 60-minute forecast after it was decayed, reflecting the physical intuition that the impact of the mutation diminishes exponentially over time—a sharp drop in irradiance after 15 minutes, and its residual effect is significantly reduced after 60 minutes.

[0074] Second item (Aging penalty): This separate item specifically addresses long-term equipment degradation. When the equipment is in perfect health... , This item is zero, with no penalty. When the equipment is severely degraded (e.g., ... )hour, The penalty takes effect. Subtract the corresponding proportion from the order of magnitude. Again, it acts as a confidence gate to prevent accidental deductions under abnormal coupling conditions. This penalty does not decay over time because equipment aging is a continuous and irreversible process, and its effects can extend to 60 minutes or even further into the future.

[0075] Two types of spatiotemporal division of labor for correction: It is primarily designed to rapidly follow physical mutations, with a response timescale in the minute range. It primarily inherits short-term trends and subtracts long-term decay, with a response timescale on the order of hours. The two are complementary and do not interfere with each other.

[0076] S45: The preliminary prediction value is corrected based on the transient correction and inertial correction to obtain the final prediction value.

[0077] (1) The correction amount is added to the initial prediction value to obtain the final prediction value: ; Will The output is sent to the EMS energy management system as the basis for power plant scheduling and operation and maintenance decisions.

[0078] Example 2 like Figure 4 As shown, this embodiment provides a dynamic power generation prediction system that integrates meteorological and equipment status data, used to implement the above-described dynamic power generation prediction method that integrates meteorological and equipment status data, including: The data acquisition module is used to simultaneously acquire meteorological and environmental data and power generation equipment status data, which are then combined into a dual-source synchronous dataset after preprocessing. The fusion module is used to extract meteorological impact features and power generation equipment status features from the dual-source synchronous dataset, perform deep fusion and correlation modeling of the dual-source data to obtain the fused feature vector, and generate a dynamic correlation weight matrix. The prediction module is used to obtain preliminary prediction values ​​by dynamically predicting power generation through a dynamic adaptive prediction model, taking the fused feature vector as input. The correction module is used to construct a dual reference system of instantaneous physical reference and inertial physical reference to quantify the degree of deviation from the actual physics, and to evaluate the credibility of the deviation using a dynamic correlation weight matrix. Based on the dual reference system and the credibility, an adaptive correction amount is generated to perform linkage correction on the preliminary prediction value to obtain the final prediction value.

[0079] Working Principle: This invention uses "dual-source fusion dynamic prediction and physical correction" as its main approach to achieve high-precision dynamic prediction of power generation. First, meteorological (irradiance, temperature, wind speed) and equipment status (attenuation rate, temperature rise rate, inverter efficiency) data are collected simultaneously. After feature extraction and cross-construction, principal component analysis (PCA) is used to decouple multicollinearity, generating a fused feature vector that allows the model to clearly distinguish the sources of power generation deviation (sudden meteorological changes or equipment aging). Second, the fused feature sequence is input into a three-layer GRU network, which simultaneously outputs preliminary predictions for the next 15 and 60 minutes. The network weights are updated in real time through online gradient descent, enabling the model to adaptively track equipment aging and meteorological changes, overcoming the limitations of fixed parameters. Finally, instantaneous and inertial dual physical benchmarks are constructed to quantify the degree of actual physical deviation. A dynamic correlation weight matrix is ​​used to extract coupling strength and temporal stability, generating a correction confidence level as a safety gate—the correction is released only when the meteorological equipment is functioning normally; otherwise, it is automatically suppressed. Based on confidence level, equipment health factors, and the intensity of physical mutations, transient corrections (for rapid, minute-level changes) and inertial corrections (for inheriting trends and subtracting long-term decays) are continuously calculated and superimposed on the initial prediction value, outputting a smooth, physically consistent final result across two time scales. The entire process has no hard cutoff threshold, ensuring stable and reliable predictions and effectively improving prediction accuracy and engineering practicality under complex operating conditions.

[0080] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0081] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for dynamic prediction of power generation integrating meteorological and equipment status data, characterized in that, Includes the following steps: Meteorological and environmental data and power generation equipment status data are collected simultaneously and combined into a dual-source synchronous dataset after preprocessing. Meteorological impact features and power generation equipment status features are extracted from the dual-source synchronous dataset. Deep fusion and correlation modeling of the dual-source data are performed to obtain the fused feature vector and generate a dynamic correlation weight matrix. Using the fused feature vector as input, a preliminary predicted value is obtained by dynamically predicting power generation through a dynamic adaptive prediction model. A dual reference system of instantaneous physical reference and inertial physical reference is constructed to quantify the degree of deviation from the actual physics. The reliability of the deviation is evaluated using a dynamic correlation weight matrix. Based on the dual reference system and the reliability, an adaptive correction amount is generated to perform a linkage correction on the preliminary prediction value to obtain the final prediction value.

2. The method for dynamic prediction of power generation integrating meteorological and equipment status as described in claim 1, characterized in that, The process of extracting meteorological impact features and power generation equipment status features from a dual-source synchronous dataset, and then performing deep fusion and correlation modeling of the dual-source data to obtain a fused feature vector, specifically includes the following steps: Extract meteorological impact feature vectors and power generation equipment status feature vectors from the dual-source synchronous dataset; Normalized cross features are obtained by cross-constructing the meteorological impact feature vector and the power generation equipment status feature vector; The original meteorological impact feature vector and the power generation equipment status feature vector are concatenated with the cross features to form the original high-dimensional feature vector; The original meteorological impact feature vector, power generation equipment status feature vector, and original high-dimensional feature vector are dynamically constructed into a covariance matrix based on their respective historical sliding windows, and principal component analysis is performed. The number of principal components is adaptively selected based on the cumulative contribution rate threshold for independent dimensionality reduction. The dimensionality reduction results of each group are then concatenated into the final fused feature vector to avoid multicollinearity among meteorological features, power generation equipment status features, and cross features, which could lead to instability in the training of the GRU model.

3. The method for dynamic prediction of power generation integrating meteorological and equipment status as described in claim 2, characterized in that, The meteorological impact characteristics include effective irradiance, ambient temperature, and wind speed fluctuation rate; the power generation equipment status characteristics include component attenuation rate, bearing temperature rise rate, and inverter efficiency deviation.

4. The method for dynamic prediction of power generation integrating meteorological and equipment status as described in claim 1, characterized in that, The generation of the dynamic correlation weight matrix refers to constructing a dynamic correlation weight matrix by calculating the sliding Pearson correlation coefficient between each component of meteorological impact characteristics and each component of power generation equipment status characteristics over a certain period of time, which serves as a statistical baseline for subsequent judgment on whether the meteorological-equipment linkage is abnormal.

5. The method for dynamic prediction of power generation integrating meteorological and equipment status as described in claim 1, characterized in that, The process of using fused feature vectors as input to dynamically predict power generation through a dynamic adaptive prediction model to obtain preliminary predicted values ​​includes the following steps: The fused feature vector at the current moment and the sequence of fused feature vectors from the previous several historical moments constitute the input sequence tensor; The input sequence tensor is expanded by time step and input into the prediction model to obtain preliminary prediction values ​​for power generation.

6. The method for dynamic prediction of power generation integrating meteorological and equipment status as described in claim 1, characterized in that, The construction of a dual reference system, consisting of an instantaneous physical reference and an inertial physical reference, to quantify the degree of deviation from the actual physical reality, and the evaluation of the reliability of the deviation using a dynamic correlation weight matrix, followed by the generation of an adaptive correction amount based on the dual reference system and the reliability to perform a linked correction on the initial prediction value to obtain the final prediction value, includes the following steps: Real-time construction of instantaneous physical reference power and inertial physical reference power as dual-scale physical reference benchmarks; Determine the intensity of physical field mutations and equipment health factors based on instantaneous physical reference power and inertial physical reference power; The corrected confidence level is determined based on the dynamic correlation weight matrix; The transient correction and inertial correction are calculated based on the instantaneous physical reference power, inertial physical reference power, physical field abrupt change intensity, equipment health factor, and correction confidence level. The initial prediction value is corrected based on the transient correction and the inertial correction to obtain the final prediction value.

7. The method for dynamic prediction of power generation integrating meteorological and equipment status as described in claim 6, characterized in that, The determination of the corrected confidence level based on the dynamic correlation weight matrix includes the following steps: The coupling strength between meteorological conditions and power generation equipment operating conditions is determined based on the dynamic correlation weight matrix; Then, the temporal stability of the linkage between meteorological and power generation equipment is obtained based on the dynamic correlation weight matrix; The corrected confidence level is obtained based on the coupling strength and timing stability.

8. A dynamic power generation prediction system integrating meteorological and equipment status data, characterized in that, The method for dynamically predicting power generation by integrating meteorological and equipment status as described in any one of claims 1 to 7 includes: The acquisition module is used to simultaneously acquire meteorological and environmental data and power generation equipment status data, which are then preprocessed and combined into a dual-source synchronous dataset. The fusion module is used to extract meteorological impact features and power generation equipment status features from the dual-source synchronous dataset, perform deep fusion and correlation modeling of the dual-source data to obtain the fused feature vector, and generate a dynamic correlation weight matrix. The prediction module is used to obtain preliminary prediction values ​​by dynamically predicting power generation through a dynamic adaptive prediction model, taking the fused feature vector as input. The correction module is used to construct a dual reference system of instantaneous physical reference and inertial physical reference to quantify the degree of deviation from the actual physics, and to evaluate the credibility of the deviation using a dynamic correlation weight matrix. Based on the dual reference system and the credibility, an adaptive correction amount is generated to perform linkage correction on the preliminary prediction value to obtain the final prediction value.

Images