New energy consumption measurement method for hydrogen production process combined with real-time data

Abstract

The present application relates to the technical field of new energy hydrogen production, and discloses a new energy consumption measurement method for hydrogen production process combined with real-time data. The method comprises obtaining hydrogen production system operation parameters and new energy power generation data, the hydrogen production system operation parameters including electrolytic cell voltage fluctuation curve and hydrogen purity detection value, and the new energy power generation data including wind power prediction deviation rate and photovoltaic actual output curve; a new energy consumption evaluation model is established, the electrolytic cell voltage fluctuation curve and the wind power prediction deviation rate are time-aligned, and a dynamic matching degree index set is generated; based on the set, a multi-scale feature extraction network is called to analyze the photovoltaic actual output curve, and a fusion analysis result containing output stability and consumption potential features is obtained; the result is input into a consumption optimization model for adaptive weight distribution, and a new energy consumption measurement value containing time dimension consumption efficiency sequence and hydrogen production equipment response parameters is output, which is suitable for the new energy consumption evaluation and scheduling demand in hydrogen production.

Description

Technical Field

[0001] This invention relates to the field of new energy hydrogen production technology, specifically a method for calculating the new energy consumption of hydrogen production processes by combining real-time data. Background Technology

[0002] In the field of integrated application of new energy power generation and hydrogen production, with the continuous expansion of the installed capacity of renewable energy sources such as wind power and photovoltaics, the volatility and intermittency of their output have become key factors restricting the stable operation of hydrogen production systems. Currently, in the hydrogen production process, the calculation of new energy consumption largely relies on static data or single-dimensional parameters, failing to fully consider the dynamic correlation between the operating status of the hydrogen production system and new energy power generation data. For example, traditional calculation methods often ignore the temporal coupling relationship between electrolyzer voltage fluctuations and wind power prediction deviations, using only average power generation data or equipment operating parameters over a fixed time period for calculation, resulting in significant deviations between the calculation results and actual operating scenarios.

[0003] Photovoltaic power output is significantly affected by factors such as sunlight intensity and weather changes, exhibiting obvious intraday and seasonal fluctuations in its output curve. Existing calculation models lack multi-scale analytical capabilities for actual photovoltaic output curves, failing to effectively extract output stability and absorption potential characteristics across different time dimensions. Consequently, it is difficult to accurately assess the actual absorption capacity of photovoltaic energy in hydrogen production. Furthermore, traditional absorption optimization models often employ preset weight allocation methods, failing to adaptively adjust to real-time changes in hydrogen production system operating parameters and renewable energy generation data. When the wind power prediction deviation rate increases or abnormal fluctuations occur in the electrolyzer voltage, the calculation results are prone to distortion, failing to provide a reliable reference for the operation and scheduling of the hydrogen production system. This not only affects the utilization rate of renewable energy resources but may also lead to frequent adjustments in the operating status of hydrogen production equipment, increasing equipment wear and operating costs.

[0004] In practical applications, while some calculation methods attempt to incorporate real-time data, they are limited to simple superposition of single-type data, failing to establish a dynamic correlation mechanism between multi-source data. This prevents the coordinated analysis of electrolyzer operating parameters, wind power data, and photovoltaic output data, resulting in a lack of systematicity and completeness in the calculation process. For example, when wind power experiences a sudden drop, failure to adjust the absorption strategy in conjunction with electrolyzer voltage fluctuations may lead to unstable energy supply during hydrogen production, affecting hydrogen production efficiency and purity. Conversely, focusing solely on electrolyzer operating parameters while ignoring dynamic changes in renewable energy generation data may result in overestimation or underestimation of renewable energy absorption, failing to match actual energy supply and demand balance requirements. These issues make current renewable energy absorption calculations in hydrogen production processes insufficient to meet the demands of large-scale, highly stable hydrogen production, hindering the industrialization and application of renewable energy hydrogen production technologies. Summary of the Invention

[0005] The purpose of this invention is to provide a method for calculating the new energy consumption of hydrogen production processes by combining real-time data, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, this invention provides a method for calculating the renewable energy consumption of hydrogen production processes by incorporating real-time data. The method includes:

[0007] The hydrogen production system operating parameters and new energy power generation data are obtained. The hydrogen production system operating parameters include the electrolyzer voltage fluctuation curve and hydrogen purity detection value. The new energy power generation data includes the wind power prediction deviation rate and the photovoltaic actual output curve.

[0008] A new energy consumption assessment model is established, and the voltage fluctuation curve of the electrolyzer and the wind power prediction deviation rate are time-series aligned to generate a dynamic matching index set.

[0009] Based on the dynamic matching degree index set, a multi-scale feature extraction network is invoked to perform segmented analysis on the actual photovoltaic output curve, generating a fusion analysis result that includes output stability features and absorption potential features.

[0010] The fusion analysis results are input into the absorption capacity optimization model for adaptive weight allocation, and the calculated value of new energy absorption capacity is output. The calculated value of new energy absorption capacity includes the absorption efficiency sequence based on the time dimension and the response parameters of hydrogen production equipment.

[0011] Preferably, the establishment of the new energy consumption assessment model includes:

[0012] Anomaly detection processing is performed on the voltage fluctuation curve of the electrolytic cell to generate voltage stability evaluation index;

[0013] The wind power prediction deviation rate is compared with a preset threshold to generate a power matching score.

[0014] A dynamic matching index set is constructed based on the voltage stability evaluation index and the power matching score. The dynamic matching index set includes timestamp-aligned voltage-power correlation parameters.

[0015] Preferably, the step of calling a multi-scale feature extraction network to perform segmented analysis of the actual photovoltaic output curve includes:

[0016] The actual photovoltaic output curve is sliced ​​according to a preset time window to generate multiple output segments;

[0017] For each output segment, high-frequency feature extraction and low-frequency trend analysis are performed simultaneously to generate multi-scale feature vectors;

[0018] The multi-scale feature vectors and the dynamic matching degree index set are spatiotemporally correlated to generate the output stability feature matrix.

[0019] Based on the similarity matching between the output stability feature matrix and historical absorption data, a probability distribution of absorption potential is generated.

[0020] Preferably, the step of inputting the fusion analysis results into the absorption capacity optimization model for adaptive weight allocation includes:

[0021] Construct a priority evaluation function for power absorption, wherein the priority evaluation function includes the variance coefficient of the output stability feature matrix and the entropy value of the probability distribution of the absorption potential;

[0022] The weight parameters of the consumption priority evaluation function are dynamically adjusted based on real-time updated new energy power generation data.

[0023] The absorption efficiency sequence of the time dimension is reordered based on the adjusted weight parameters to generate an optimized absorption amount measurement value.

[0024] Preferably, the method further includes:

[0025] Real-time monitoring of the hydrogen production system's operating status, collecting real-time operating data including the rate of change in electrolysis efficiency and the energy consumption of the cooling system;

[0026] The real-time operating data is cross-validated with the calculated value of new energy consumption to generate a system compatibility assessment result.

[0027] When the system compatibility assessment result exceeds the preset range, the parameter update mechanism of the absorption capacity optimization model is triggered.

[0028] Preferably, the parameter update mechanism for triggering the absorption capacity optimization model includes:

[0029] Extract the abnormal fluctuation characteristics of new energy power generation data within a preset time range before and after the current moment;

[0030] The historical case matching engine is invoked to find historical scene data similar to the abnormal fluctuation characteristics.

[0031] The weight parameters of the absorption priority evaluation function are corrected based on the matched historical scenario data;

[0032] The corrected weight parameters are injected into the absorption optimization model to recalculate the measured value of new energy absorption.

[0033] Preferably, the method further includes:

[0034] A multi-source data verification mechanism is established, which includes deviation analysis between new energy prediction data and actual power output data.

[0035] When a data deviation is detected to exceed a threshold, a data compensation algorithm is activated to smooth the actual photovoltaic output curve.

[0036] The smoothed data is then re-input into the multi-scale feature extraction network for feature extraction, and the fusion analysis results are updated.

[0037] Preferably, the startup data compensation algorithm includes:

[0038] Identify the location and magnitude of abrupt changes in the actual photovoltaic output curve;

[0039] A local compensation function is constructed based on the output change trend in adjacent time periods;

[0040] A sliding window approach is used to progressively correct the abrupt change points, generating a continuous and smooth output curve.

[0041] Record the timestamps and correction amounts of all compensation operations to generate a data compensation log.

[0042] Preferably, the method further includes:

[0043] A cross-platform data adaptation interface is constructed, which supports the input of new energy monitoring data in different formats;

[0044] The input monitoring data is processed in a unified and standardized manner to generate a standardized data stream with time synchronization characteristics;

[0045] The standardized data stream is time-aligned with the operating parameters of the hydrogen production system to ensure the consistency of data processing timing.

[0046] Preferably, the standardized processing of the input monitoring data includes:

[0047] Extract the collection time information and data precision parameters from different data sources;

[0048] Establish a time reference axis and align the timestamps of all monitoring data to the time reference axis;

[0049] An interpolation algorithm is set based on the differences in data precision to supplement missing data. The supplemented data is then normalized to generate the standardized data stream.

[0050] Compared with the prior art, the beneficial effects of the present invention are:

[0051] By comprehensively acquiring operating parameters of the hydrogen production system and data from new energy power generation, this method overcomes the limitations of traditional calculation methods that rely on single or incomplete data. It incorporates electrolyzer voltage fluctuation curves and hydrogen purity test values ​​into the operating parameters of the hydrogen production system, while also covering new energy power generation data such as wind power prediction deviation rates and actual photovoltaic output curves. This achieves collaborative acquisition of real-time data from multiple sources, providing a comprehensive and accurate data foundation for subsequent calculations and avoiding calculation errors caused by missing or isolated data.

[0052] In the model building phase, this method establishes a new energy consumption assessment model and performs time-series alignment processing on the electrolyzer voltage fluctuation curve and the wind power prediction deviation rate to generate a dynamic matching degree index set. This effectively captures the dynamic correlation between the operating status of hydrogen production equipment and wind power generation data. This time-series alignment processing can reflect the changing coupling characteristics of the two in different time periods. For example, during periods when the wind power prediction deviation rate increases, the fluctuation of the electrolyzer voltage is analyzed simultaneously, thereby more accurately identifying the key factors affecting new energy consumption. Compared with the static analysis that ignores time-series correlation in traditional methods, this significantly improves the fit between the calculation process and the actual operating scenario.

[0053] By employing a multi-scale feature extraction network based on a dynamic matching degree index set to segment and analyze the actual photovoltaic (PV) output curve, the network can deeply explore the changing patterns of PV output at different time scales. The multi-scale feature extraction network can perform hierarchical analysis of intraday short-term fluctuations, diurnal trends, and seasonal periodic characteristics of the PV output curve, generating a fusion analysis result that includes output stability features and absorption potential features. This not only identifies the stable operating range of PV output but also accurately assesses the absorption capacity of PV energy in different time periods, solving the problem that traditional models cannot effectively analyze the multi-dimensional characteristics of PV output and providing a more targeted analytical basis for subsequent absorption optimization.

[0054] The fusion analysis results are input into the energy consumption optimization model for adaptive weight allocation, enabling the model to dynamically adjust the weight ratio of each influencing factor based on real-time data changes. For example, when the wind power prediction deviation rate increases, the model can automatically increase the weight of the electrolyzer voltage fluctuation curve in the calculation to better match the unstable characteristics of wind power generation; while when photovoltaic output enters a stable range, the weight of the output stability characteristics can be appropriately adjusted to ensure that the calculation results can respond in real time to changes in the operating status of new energy power generation and hydrogen production systems. This adaptive adjustment method avoids the calculation distortion problem of traditional preset weight models in scenarios with dynamically changing data, improving the flexibility and accuracy of the calculation results.

[0055] The final output of the calculated renewable energy consumption value includes a time-based consumption efficiency sequence and hydrogen production equipment response parameters. It not only presents the changes in renewable energy consumption efficiency over different time periods but also provides operational response data for the hydrogen production equipment during corresponding periods, such as voltage adjustment amplitude and operating power changes in the electrolyzer. This output format provides more comprehensive reference information for the operation and scheduling of the hydrogen production system. Schedulers can rationally arrange the access periods for renewable energy based on the consumption efficiency sequence and optimize equipment operation strategies by combining the hydrogen production equipment response parameters. This achieves both efficient consumption of renewable energy resources and stable operation of the hydrogen production equipment, reducing equipment wear and operating costs, and providing strong support for the large-scale application of renewable energy hydrogen production technology.

[0056] This method organically combines multi-source data collaboration, time-series alignment processing, multi-scale analysis, and adaptive weight allocation to construct a complete dynamic calculation system that can effectively address the volatility of new energy power generation and the complexity of hydrogen production system operation. In practical applications, regardless of short-term sudden changes in wind power output, seasonal fluctuations in photovoltaic output, or abnormal fluctuations in electrolyzer voltage, this method can maintain the accuracy and reliability of the calculation results through dynamic analysis of real-time data and adaptive adjustment of the model. This avoids the limitations of traditional calculation methods under complex operating conditions and further improves the overall operational efficiency of new energy hydrogen production systems. Attached Figure Description

[0057] Figure 1 This is a schematic diagram illustrating the working principle of the new energy consumption calculation method for hydrogen production processes that incorporate real-time data, as described in this invention.

[0058] Figure 2 The flowchart for segmented parsing of a multi-scale feature extraction network;

[0059] Figure 3 A flowchart for optimizing the parameter update mechanism of the model for the amount of wastewater absorbed. Detailed Implementation

[0060] 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.

[0061] Please see Figure 1This invention provides a method for calculating the renewable energy consumption of a hydrogen production process that incorporates real-time data. The method includes: achieving accurate calculation by acquiring operating parameters of the hydrogen production system and renewable energy power generation data. The operating parameters of the hydrogen production system include the voltage fluctuation curve of the electrolyzer and the hydrogen purity detection value. The renewable energy power generation data includes the wind power prediction deviation rate and the actual output curve of the photovoltaic system. A renewable energy consumption assessment model is established, and the voltage fluctuation curve of the electrolyzer and the wind power prediction deviation rate are time-series aligned to generate a dynamic matching degree index set. Based on the dynamic matching degree index set, a multi-scale feature extraction network is invoked to perform segmented analysis on the actual output curve of the photovoltaic system, generating a fusion analysis result that includes output stability features and consumption potential features. The fusion analysis result is input into a consumption optimization model for adaptive weight allocation, and the calculated renewable energy consumption value is output. The calculated renewable energy consumption value includes a time-dimensional consumption efficiency sequence and hydrogen production equipment response parameters.

[0062] Example 1: The establishment of the new energy consumption assessment model involves the detection and processing of outliers in the voltage fluctuation curve of the electrolyzer. A sliding window variance calculation algorithm is used to scan the voltage sequence. The window size is set according to the sampling frequency and equipment response characteristics, with fifty data points as one calculation unit. Each data point represents a millisecond-level voltage measurement value. The algorithm uses an overlapping sliding mechanism to enhance detection sensitivity when traversing the entire voltage sequence. When the voltage of three consecutive sampling points deviates from the mean by more than a certain multiple of the standard deviation, it is marked as an outlier. This multiple threshold is determined based on the statistical distribution characteristics of historical normal operation data. Outlier detection not only considers instantaneous deviation but also combines the changing trend of the preceding and following data points for comprehensive judgment. After removing outliers, the voltage curve is reconstructed through an interpolation algorithm. The choice of interpolation method considers the electrochemical response characteristics of the electrolyzer and the continuity requirements of the voltage signal. The reconstructed curve is used to generate a voltage stability assessment index. This index includes multiple dimension parameters such as the voltage fluctuation coefficient and the proportion of continuous stable time. The fluctuation coefficient reflects the degree of voltage deviation from the benchmark value, while the proportion of stable time reflects the proportion of time the system maintains stable operation.

[0063] When comparing the wind power prediction deviation rate with the preset threshold, the preset threshold is not a fixed value but is dynamically adjusted based on the historical operating data of the wind farm. The adjustment strategy takes into account the seasonality of wind speed changes and the daily time period characteristics so that the threshold can adapt to different operating conditions. The threshold range is extracted from the historical data using the percentile statistical method to extract the upper and lower bounds of the normal fluctuation range. The process of generating the power matching score is a multi-factor weighted calculation. The score calculation is based on the cumulative duration of the deviation rate falling within the threshold range and the deviation magnitude weighted value. The duration ratio reflects the persistence of the power prediction and the actual matching, while the deviation magnitude weighted value quantifies the severity of the mismatch. The scoring algorithm uses normalization processing to make the output result between zero and one, which is convenient for subsequent model use. A dynamic matching index set was constructed based on voltage stability assessment indicators and power matching degree scores. Each data point in the set includes a timestamp accurate to the millisecond level to ensure the accuracy of time sequence alignment. The timestamp alignment process adopted a high-precision time synchronization protocol to coordinate the acquisition devices of different data sources. The voltage-power correlation parameters were calculated by normalizing the product after timestamp alignment. A sliding time window was introduced in the calculation process to capture dynamic change characteristics. Normalization eliminated dimensional differences, making different indicators comparable. The final index set is a data sequence with strict time sequence correlation. This sequence provides a quantitative description of the voltage and power matching relationship for subsequent analysis. The calculation of voltage stability assessment indicators also considered the thermal inertia and electrochemical response delay characteristics of the electrolyzer. A time delay correction algorithm was introduced in the index generation process to accurately reflect the real state of the system. The fluctuation coefficient was calculated using a detrended fluctuation analysis method to distinguish between normal and abnormal fluctuations. The statistics of the stable duration percentage were based on the length of time the voltage value remained within the rated operating range. These indicators together constitute a comprehensive evaluation system for the operational stability of the electrolyzer.

[0064] The dynamic threshold adjustment mechanism for power matching score is updated hourly to ensure real-time performance. The update process uses a sliding window approach, taking data from the most recent 24 hours for statistical calculation. A robust estimation method is employed to set the threshold range to avoid the influence of extreme values. The weighting coefficients in the scoring calculation are adaptively adjusted based on the duration of the deviation; deviations with longer durations receive higher weighting coefficients. This design makes the scoring results more reflective of the actual matching status during operation. Data quality checks were also performed during the construction of the dynamic matching index set, including data integrity verification and outlier review. Timestamp alignment uses a bidirectional time synchronization algorithm, taking into account network transmission latency and device clock errors. The calculation of voltage-power correlation parameters incorporates a sliding correlation coefficient to capture the dynamic correlation strength. Normalization uses a minimum-maximum scaling method to transform the data to the zero-to-one interval. The final index set is stored as a time-series array for easy access by subsequent modules. The entire process of establishing the new energy consumption assessment model adopts a modular design concept. Data is transmitted between various calculation modules through standard data interfaces. The modules are equipped with a data caching mechanism to cope with the unevenness of data flow. During the calculation process, running logs are generated in real time to record intermediate results and abnormal situations. The log information includes data processing time, parameter values, and calculation status. This design facilitates subsequent model optimization and fault diagnosis.

[0065] Example 2: See Figure 2 The multi-scale feature extraction network segments the actual photovoltaic power output curve according to preset time windows. The time window settings comprehensively consider the characteristics of photovoltaic power generation and data analysis needs: a 15-minute window captures rapid fluctuations caused by cloud cover, an hourly window reflects the trend of solar radiation intensity changes, and a daily window corresponds to the overall macroscopic changes in power generation. Each window uses an overlapping slicing method to enhance data continuity, with an overlap rate of 30% between windows to ensure a smooth transition in feature extraction. After slicing generates multiple power output segments, the feature extraction stage begins. High-frequency feature extraction and low-frequency trend analysis are performed simultaneously on each power output segment. High-frequency feature extraction uses a discrete wavelet transform algorithm to decompose minute-level fluctuation components, with the wavelet basis function chosen considering the abrupt changes in the photovoltaic power output signal. Low-frequency trend analysis uses a weighted moving average method to extract hourly variation patterns, with weight coefficients distributed exponentially based on time distance. The generation process of the multi-scale feature vector includes a feature fusion step, combining feature values ​​from different scales into a vector of a unified dimension. The vector dimension corresponds to the number of time windows for subsequent processing.

[0066] A spatiotemporal correlation calculation is performed between multi-scale feature vectors and a dynamic matching degree index set. This calculation employs a distributed computing framework to improve processing efficiency. The Pearson correlation coefficient calculates the linear correlation between the feature vectors and the dynamic matching degree index, while the spatiotemporal delay mutual information captures nonlinear correlation characteristics. A sliding window approach is used to progressively analyze the correlation characteristics across different time periods. The generation of the power output stability feature matrix includes matrix normalization, which ensures comparability of feature values ​​across different time periods through row normalization. The matrix rows correspond to time segment sequences, and columns include multiple dimensions such as fluctuation amplitude, correlation, and delay parameters. Similarity matching is performed between the power output stability feature matrix and historical absorption data, sourced from a database of long-term hydrogen production system operation records. An improved dynamic time warping algorithm is used for similarity matching, introducing constraints to improve computational efficiency. The matching process calculates the similarity score between the current matrix and each pattern in the historical pattern library, considering both shape similarity and numerical closeness. The generation of the absorption potential probability distribution uses kernel density estimation, where the probability value represents the maximum absorption probability achievable under current photovoltaic power output conditions. The probability distribution curve is smoothed to avoid overfitting. When the fusion analysis results are input into the absorption capacity optimization model for adaptive weight allocation, the construction of the absorption priority evaluation function involves multiple parameter optimization steps. The variance coefficient in the function expression reflects the degree of power fluctuation, which is obtained by calculating the coefficient of variation of each row vector of the feature matrix. The Shannon entropy value characterizes the absorption uncertainty and is obtained by calculating the information entropy based on the probability distribution. The function parameters are initialized using the historical best value as the starting point. The real-time updated new energy power generation data comes from the data flow interface of the monitoring system. The data update frequency is synchronized with the feature extraction. The dynamic adjustment process adopts an online learning mechanism. The weight parameter adjustment is based on solving the optimal weight combination using the gradient descent method. The learning rate setting adopts an adaptive adjustment strategy, and the weight coefficient is updated every five minutes to ensure real-time performance.

[0067] The time-dimensional absorption efficiency sequence is reordered based on adjusted weight parameters. The sorting algorithm employs a weighted scoring mechanism, with the score calculated as a linear combination of weight parameters and feature values. The sorting process considers time continuity constraints to avoid drastic jumps. The optimized absorption capacity calculation includes absorption priority scores for each time period, and the scores are normalized to between zero and one hundred for easy understanding. The response parameters of the hydrogen production equipment are automatically generated based on the score values, and the parameter settings take into account equipment operating constraints and safety boundaries. The final output calculation format adopts a standardized data protocol, including fields such as timestamp, absorption capacity, and equipment parameters. The entire implementation process adopts a pipeline architecture design, with each processing stage decoupled through a data buffer. Processing nodes have fault tolerance mechanisms that can automatically restart when an anomaly occurs in a stage. The data flow monitoring system tracks processing progress and resource usage in real time, triggering an early warning mechanism and recording detailed logs in case of anomalies. The system's operating status is displayed through a visual interface for monitoring personnel to easily grasp the real-time situation.

[0068] Taking a wind-solar hybrid power generation project in a coastal area encountering alternating weather conditions on June 21, 2023 as an example, the morning saw rapidly moving fragmented clouds causing frequent fluctuations in the photovoltaic power station's output. In the afternoon, the clouds gradually dissipated, resulting in stable sunshine, while the local wind farm experienced stronger sea breezes at midday. In this scenario, the photovoltaic power station had a rated capacity of 100MW. The actual output curve recorded minute-by-minute power generation data from 08:00 to 16:00. The data acquisition system detected 17 power drop events exceeding 20% ​​of the rated capacity between 09:45 and 10:30, with the largest drop reaching 65% of the rated capacity. When processing the daily output curve using a multi-scale feature extraction network, it first sliced ​​the data according to three preset time windows: 15 minutes, 1 hour, and 4 hours. The 15-minute window focused on minute-level fluctuations caused by cloud movement, the 1-hour window analyzed the trend of solar intensity changes, and the 4-hour window grasped the overall pattern of power generation changes. Each power output segment undergoes simultaneous high-frequency feature extraction and low-frequency trend analysis. High-frequency analysis uses the Db4 wavelet basis function to decompose the fluctuation components at a 5-minute scale. Low-frequency trend analysis uses a weighted moving average method to extract hourly variation features. The generated multi-scale feature vector contains 128 eigenvalues ​​to characterize power output characteristics at different time scales. The dynamic matching degree index set includes the daily wind farm power prediction deviation rate data, with a maximum deviation rate reaching 28% during the period from 11:00 to 13:00, characterized by a sustained positive deviation. This data is timestamped and aligned with the electrolyzer voltage fluctuation curve, achieving millisecond-level alignment accuracy. Spatiotemporal correlation calculation employs a distributed computing framework to calculate the mutual information and correlation coefficient between photovoltaic power output characteristics and wind power deviation rate within each time segment. The generated power output stability feature matrix has rows corresponding to 288 time segments (at 5-minute intervals) and columns containing 12 feature dimensions, including fluctuation amplitude, correlation coefficient, and time delay parameters. When performing similarity matching based on the power output stability feature matrix and historical absorption data, the system retrieves operational records of the same weather pattern in June of the past three years from the database and uses a dynamic time warping algorithm to calculate the similarity score between the current matrix and historical patterns. The matching results show that the current fluctuation pattern has 82% similarity to the weather characteristics of June 15th of last year, when the absorption rate reached a historical high of 91%. Based on this, the generated absorption potential probability distribution shows that the current period has an absorption probability of 85%-92%. After the fusion analysis results are input into the absorption capacity optimization model, the absorption priority evaluation function begins to calculate the variance coefficient of the feature matrix and the entropy value of the probability distribution. The variance coefficient calculation shows that the fluctuation intensity during the period from 09:00 to 10:30 exceeds the normal range by 3.2 times, and the entropy value analysis indicates that the absorption uncertainty during this period is 40% higher than the average level. The dynamic adjustment module for weight parameters recalculates the weight allocation scheme every 5 minutes based on the real-time updated photovoltaic output data. During the 10:25 adjustment cycle, the weight coefficient of the volatility index is increased from 0.3 to 0.45, while the weight ratio of other factors is reduced accordingly.Based on the adjusted weight parameters, the absorption efficiency sequence was reordered. The system identified the 11:30-13:00 period as having the optimal absorption priority score, reaching 92 points, while the 09:45-10:30 period only scored 68 points. The optimized absorption calculation value generated a time-dimensional absorption efficiency sequence, showing that a 95% absorption rate could be achieved during the midday period, while it was recommended to control it at around 82% during morning fluctuations. The hydrogen production equipment response parameters were recommended to maintain 80% of rated power during fluctuations to cope with the risk of sudden power drops. The entire processing generated 287 feature extraction records and 156 weight adjustment operations. System logs showed that multi-scale feature extraction took 3.7 seconds, spatiotemporal correlation calculation took 2.1 seconds, and the optimized model processing cycle was 1.8 seconds. The final output absorption calculation value was transmitted to the hydrogen production system control system via a data interface. Actual operation data showed that the daily absorption rate reached 89.7%, with a deviation from the predicted value within 2%.

[0069] Example 3: See Figure 3 Real-time monitoring of the hydrogen production system's operating status is achieved through a sensor network deployed on the electrolyzer and auxiliary equipment. Data acquisition is set to once per second to ensure the capture of rapidly changing operating parameters. The electrolysis efficiency rate of change is calculated based on the ratio of real-time hydrogen production to electrical energy consumption, and the instantaneous change is obtained using the finite difference method. Cooling system energy consumption data comes from the power consumption of the thermal management unit recorded by the power metering module. All real-time operating data is transmitted to the central processing unit via an Industrial Internet of Things (IIoT) protocol and timestamped. Data stream processing employs a first-in-first-out (FIFO) buffer mechanism to address transmission delays and out-of-order issues. When cross-validating real-time operating data with calculated new energy consumption values, a time-synchronized data comparison window is established. The window length is set to a five-minute sliding interval based on the system response time. Residual analysis calculates the deviation using the following formula: in: This represents the average relative deviation, where N is the number of data points and M is the mean relative deviation. i A represents the i-th calculated value. i This represents the i-th actual measured value. To prevent small constants from being divided by zero, the generation of system compatibility assessment results includes multi-level judgment logic. First, the deviation of the main parameters is calculated, and then a comprehensive score is given in combination with auxiliary indicators. When the score result exceeds the preset range determined by the equipment safety operation specifications, an update mechanism is immediately triggered.

[0070] The parameter update mechanism uses multi-level thresholds to differentiate between deviations of varying severity. Minor deviations are only logged, while significant deviations immediately trigger a model update. When extracting abnormal fluctuation characteristics from new energy power generation data, a sliding time window analysis of the most recent 30-minute data sequence is employed. Feature extraction includes multi-dimensional quantification of fluctuation amplitude, rate of change, and duration. The identification of abnormal fluctuation characteristics is based on... - The statistical detection method is combined with an expert rule base for comprehensive judgment. The historical case matching engine uses an improved k-nearest neighbor algorithm to search for similar scenes in the historical database. The algorithm introduces dynamic time-normalized distance as a similarity metric to handle the scaling and deformation of time series. The matching process prioritizes the similarity of time features and then compares the consistency of fluctuation patterns. It retrieves the most similar scene cases from the historical database and extracts the corresponding processing parameters. The matching results undergo confidence testing to ensure the reliability of the reference value. When correcting the weight parameters of the absorption priority evaluation function based on the matched historical scene data, a weighted fusion strategy is used to combine the historically optimal parameters with the current parameters. A forgetting factor is introduced during the correction process to make the system pay more attention to the recent data change trends. The adjustment range of the weight parameters is dynamically determined according to the matching degree of historical cases and the current deviation degree. The correction algorithm ensures the smoothness of parameter changes and avoids system oscillations.

[0071] When injecting the revised weight parameters into the renewable energy consumption optimization model, a hot update method is used to ensure continuous system operation. The model recalculation process employs an incremental update algorithm to improve computational efficiency. The recalculated renewable energy consumption values ​​undergo consistency checks to ensure their matching with real-time operational data. Updated model parameters and calculation results are recorded in a version management database for easy traceability and analysis. The entire real-time monitoring and parameter update process forms a closed-loop control system. The data acquisition module continuously transmits real-time operational status information, the analysis module dynamically assesses system compatibility, and the update mechanism ensures the model always adapts to actual operating conditions. The system design considers contingency plans for various abnormal situations, including communication interruptions, data anomalies, and equipment failures. The log system records in detail the time, reason, and processing result of each update trigger. These records provide a data foundation for system optimization and also meet operational audit requirements. Special attention is paid to the processing accuracy of time-series data during implementation. All data acquisition devices use a unified time synchronization protocol to eliminate clock deviations. Data analysis considers and compensates for transmission delays from different devices. Model updates are selected during periods of relatively stable system load to minimize the impact on operation. The entire mechanism design fully reflects a balance between real-time performance, accuracy, and reliability.

[0072] Example 4: The establishment of a multi-source data verification mechanism is based on continuous comparison and analysis of new energy forecast data and actual output data. This mechanism adopts a dual-channel data stream processing architecture to process forecast data and measured data separately. The deviation analysis calculation module calculates the mean absolute percentage error and root mean square error in real time. The error threshold is set with reference to the statistical distribution characteristics of historical operating data and with appropriate safety margin. When both indicators exceed the threshold for three consecutive sampling periods, the data compensation process is triggered. The data compensation algorithm requires a two-level confirmation mechanism to prevent false triggering. The first level detects abnormal fluctuations and starts the transient analysis module to determine the duration. The second level eliminates false anomalies caused by sensor failures by associating with equipment status data. After confirming that compensation is needed, the system automatically selects an appropriate compensation strategy and generates a processing log. The identification of abrupt change points in the actual photovoltaic power output curve employs a multi-scale sliding window differential algorithm. The window size is set to three levels—1 minute, 5 minutes, and 15 minutes—based on the variation pattern of solar irradiance. Determining abrupt change points requires simultaneously satisfying both amplitude and gradient thresholds. The amplitude threshold is set as a percentage of the historical maximum fluctuation range, while the gradient threshold limits the rate of change per unit time. The identification results include precise temporal location and deviation quantification of the abrupt change point. The construction of the local compensation function is based on the analysis of power output change trends in adjacent time periods before and after the abrupt change point. Trend analysis uses a weighted least squares method to fit the normal trajectory of power output change. The design principle of the compensation function is to maintain the continuity of the first derivative of the power output curve and its physical rationality. The function parameters are adaptively adjusted according to the type of abrupt change: a gradual recovery function is used for sudden drops, and a smooth decay function is used for sudden increases. All compensation functions undergo energy conservation verification to ensure that the total power generation remains consistent before and after compensation.

[0073] The sliding window progressive correction employs an iterative processing method with variable step sizes. The window width is automatically adjusted according to the magnitude of the abrupt change and remains consistent with the window size used in the identification stage. The correction process consists of three steps: first, high-intensity correction is performed in the central region of the abrupt change; second, medium-intensity smoothing is performed in the transition region; and finally, weak correction is performed in the edge region to ensure a natural transition with normal data. Each correction operation records the correction amount, timestamp, and type of compensation function used. These records form a complete data compensation log for subsequent auditing and analysis. The data compensation log uses a structured storage format and includes two parts: a time-series index and operation records. The log records include the original data value, the compensated data value, the compensation algorithm version number, the operator identifier (automatic system operations are marked as AUTO), and the compensation reason code. The compensation reason code is categorized and coded according to the anomaly type for easy statistical analysis and pattern recognition. Refer to Table 1 for a detailed record of a data compensation process.

[0074] Table 1: Photovoltaic Output Data Compensation Operation Log Record

[0075]

[0076] When the smoothed data is re-input into the multi-scale feature extraction network, a data pipeline processing mode is adopted. The re-extracted features are compared with the original features to ensure that the compensation operation does not introduce new distortions. The feature extraction parameters are kept in configuration completely consistent with the initial processing, including wavelet basis function type and moving average window size. The updated fusion analysis results are labeled with the data source version to track the processing history. A comprehensive quality control system is established throughout the data compensation process. Each compensation operation generates reverse verification data to verify the compensation effect. Verification methods include horizontal comparison with data from adjacent photovoltaic units and vertical comparison with historical data from the same period. When the verification finds that the compensation result does not meet expectations, the system automatically initiates a secondary compensation process. All compensation operations are recorded in a distributed database and synchronized to backup nodes to ensure data security. Special attention is paid to the transparency and traceability of the compensation algorithm during implementation. All compensation parameters and algorithm logic are open to system administrators for auditing. Compensation logs regularly generate statistical analysis reports to help improve the identification algorithm and compensation strategy. The system is equipped with a compensation effect evaluation module to continuously monitor the quality indicators of the data after compensation. These designs ensure the reliability and credibility of the data compensation process.

[0077] Example 5: The cross-platform data adaptation interface design adopts a modular architecture to support the input of various new energy monitoring data formats. The core of the interface includes three functional modules: a protocol parsing layer, a data conversion layer, and a unified output layer. The protocol parsing layer has built-in processing engines for OPCUA, IEC61850, and a custom TCP protocol. Each protocol engine runs independently and exchanges data through an asynchronous message queue. The data conversion layer maps the parsed raw data into a unified internal data model. The model definition includes four basic fields: measurement point identifier, data value, timestamp, and quality code. The unified output layer assembles data packets according to a standardized format and distributes them externally through a message middleware. When performing unified standardization processing on the input monitoring data, the acquisition time information and data precision parameters of different data sources are extracted first. The time information parsing takes into account special cases such as time zone conversion and leap second adjustment. The precision parameters record the effective number of digits and measurement unit of the raw data. The data quality code is automatically generated according to the status of the acquisition equipment and includes multiple levels such as normal, suspicious, out of limit, and fault. The time reference axis is established using International Atomic Time (IAT) as the reference time source, and synchronization with the time server is maintained via the NTP protocol. Transmission delay and device clock offset are calculated when converting the timestamps of all monitored data to the same time reference, with time alignment accuracy controlled within milliseconds. An adaptive strategy is adopted when setting interpolation algorithms based on data precision differences. Piecewise polynomial interpolation is used for high-precision data to preserve data characteristics, while linear interpolation is used for low-precision data to improve processing efficiency. Missing data is supplemented based on time series prediction algorithms, considering data autocorrelation and the influence of external factors. Normalization processing employs a dynamic range adjustment method, calculating scaling factors in real time based on the statistical characteristics of the data stream to generate a standardized data stream containing complete metadata information for use in subsequent processing stages.

[0078] In the specific implementation process, a wind farm's SCADA system uses the IEC 61850 protocol to transmit data. Data packets include fields such as timestamps, wind speed, power values, and equipment status. The protocol parsing layer identifies the MMS message structure and extracts data values. The data conversion layer converts wind speed units to meters per second, power values ​​to per-unit values, and timestamps to ISO 8601 format. Simultaneously, a photovoltaic monitoring system transmits string-level power generation data via the OPCUA protocol. The data includes string-formatted timestamps and floating-point power values. The parsing layer needs to process time stamps from different time zones and convert them to a unified time base. During time alignment, it was found that the wind turbine data acquisition cycle was 3 seconds while the photovoltaic data acquisition cycle was 5 seconds. The system automatically calculates the least common multiple to determine a 15-second resampling interval and uses the Lagrange interpolation algorithm to generate equally spaced data sequences. For data precision processing, wind turbine power values ​​are retained to two decimal places while photovoltaic data is retained to one. The system uniformly processes the data to two decimal places according to the higher precision. During normalization, power values ​​are divided by rated capacity to convert to per-unit values, and wind speed data is normalized using the maximum wind speed value. All data conversion coefficients are recorded in the metadata. Standardized data stream output uses the Apache Avro serialization format, including schema definitions and data content. The schema clearly defines the name, type, and unit of measurement for each field, and the data content is arranged in chronological order to form a continuous data stream. When aligning with the hydrogen production system's operating parameters, sliding window correlation analysis is used to calculate the time offset, and data cache latency is dynamically adjusted to ensure that the time deviation of all data is less than half of the processing cycle. During implementation, a data quality monitoring dashboard is established to display the access status and processing quality of each data source in real time. Data anomalies are automatically triggered, and detailed logs are recorded. The system periodically generates data quality reports, statistically analyzing indicators such as packet loss rate, latency distribution, and data processing accuracy. All processing algorithms are containerized for deployment, supporting dynamic expansion and canary upgrades to ensure continuous and stable system operation.

[0079] It should be noted that, in this document, relational terms such as "first" and "second" are used only 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 process, method, article, or apparatus.

[0080] 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 calculating the renewable energy consumption of a hydrogen production process that incorporates real-time data, characterized in that, include: The hydrogen production system operating parameters and new energy power generation data are obtained. The hydrogen production system operating parameters include the electrolyzer voltage fluctuation curve and hydrogen purity detection value. The new energy power generation data includes the wind power prediction deviation rate and the photovoltaic actual output curve. A new energy consumption assessment model is established, and the voltage fluctuation curve of the electrolyzer and the wind power prediction deviation rate are time-series aligned to generate a dynamic matching index set. Based on the dynamic matching degree index set, a multi-scale feature extraction network is invoked to perform segmented analysis on the actual photovoltaic output curve, generating a fusion analysis result that includes output stability features and absorption potential features. The fusion analysis results are input into the absorption capacity optimization model for adaptive weight allocation, and the calculated value of new energy absorption capacity is output. The calculated value of new energy absorption capacity includes the absorption efficiency sequence based on the time dimension and the response parameters of hydrogen production equipment.

2. The method for calculating the new energy consumption of hydrogen production processes combined with real-time data as described in claim 1, characterized in that, The establishment of the new energy consumption assessment model includes: Anomaly detection processing is performed on the voltage fluctuation curve of the electrolytic cell to generate voltage stability evaluation index; The wind power prediction deviation rate is compared with a preset threshold to generate a power matching score. A dynamic matching index set is constructed based on the voltage stability evaluation index and the power matching score. The dynamic matching index set includes timestamp-aligned voltage-power correlation parameters.

3. The method for calculating the new energy consumption of hydrogen production processes combined with real-time data as described in claim 2, characterized in that, The step of using a multi-scale feature extraction network to perform segmented analysis of the actual photovoltaic power output curve includes: The actual photovoltaic output curve is sliced ​​according to a preset time window to generate multiple output segments; For each output segment, high-frequency feature extraction and low-frequency trend analysis are performed simultaneously to generate multi-scale feature vectors; The multi-scale feature vectors and the dynamic matching degree index set are spatiotemporally correlated to generate the output stability feature matrix. Based on the similarity matching between the output stability feature matrix and historical absorption data, a probability distribution of absorption potential is generated.

4. The method for calculating the new energy consumption of hydrogen production processes combined with real-time data as described in claim 3, characterized in that, The step of inputting the fusion analysis results into the absorption capacity optimization model for adaptive weight allocation includes: Construct a priority evaluation function for power absorption, wherein the priority evaluation function includes the variance coefficient of the output stability feature matrix and the entropy value of the probability distribution of the absorption potential; The weight parameters of the consumption priority evaluation function are dynamically adjusted based on real-time updated new energy power generation data. The absorption efficiency sequence of the time dimension is reordered based on the adjusted weight parameters to generate an optimized absorption amount measurement value.

5. The method for calculating the new energy consumption of hydrogen production processes combined with real-time data as described in claim 4, characterized in that, Also includes: Real-time monitoring of the hydrogen production system's operating status, collecting real-time operating data including the rate of change in electrolysis efficiency and the energy consumption of the cooling system; The real-time operating data is cross-validated with the calculated value of new energy consumption to generate a system compatibility assessment result. When the system compatibility assessment result exceeds the preset range, the parameter update mechanism of the absorption capacity optimization model is triggered.

6. The method for calculating the new energy consumption of hydrogen production processes combined with real-time data as described in claim 5, characterized in that, The parameter update mechanism that triggers the absorption capacity optimization model includes: Extract the abnormal fluctuation characteristics of new energy power generation data within a preset time range before and after the current moment; The historical case matching engine is invoked to find historical scene data similar to the abnormal fluctuation characteristics. The weight parameters of the absorption priority evaluation function are corrected based on the matched historical scenario data; The corrected weight parameters are injected into the absorption optimization model to recalculate the measured value of new energy absorption.

7. The method for calculating the new energy consumption of hydrogen production processes combined with real-time data as described in claim 6, characterized in that, Also includes: A multi-source data verification mechanism is established, which includes deviation analysis between new energy prediction data and actual power output data. When a data deviation is detected to exceed a threshold, a data compensation algorithm is activated to smooth the actual photovoltaic output curve. The smoothed data is then re-input into the multi-scale feature extraction network for feature extraction, and the fusion analysis results are updated.

8. The method for calculating the new energy consumption of hydrogen production processes combined with real-time data as described in claim 7, characterized in that, The startup data compensation algorithm includes: Identify the location and magnitude of abrupt changes in the actual photovoltaic output curve; A local compensation function is constructed based on the output change trend in adjacent time periods; A sliding window approach is used to progressively correct the abrupt change points, generating a continuous and smooth output curve. Record the timestamps and correction amounts of all compensation operations to generate a data compensation log.

9. The method for calculating the new energy consumption of hydrogen production processes combined with real-time data as described in claim 8, characterized in that, Also includes: A cross-platform data adaptation interface is constructed, which supports the input of new energy monitoring data in different formats; The input monitoring data is processed in a unified and standardized manner to generate a standardized data stream with time synchronization characteristics; The standardized data stream is time-aligned with the operating parameters of the hydrogen production system to ensure the consistency of data processing timing.

10. The method for calculating the new energy consumption of hydrogen production processes combined with real-time data as described in claim 9, characterized in that, The standardized processing of the input monitoring data includes: Extract the collection time information and data precision parameters from different data sources; Establish a time reference axis and align the timestamps of all monitoring data to the time reference axis; An interpolation algorithm is set based on the differences in data precision to supplement missing data. The supplemented data is then normalized to generate the standardized data stream.

Images