A method for identifying the operating conditions of electrical load in thermal power units

Abstract

This disclosure provides a method for identifying the operating conditions of thermal power units based on electrical load. By combining image contour extraction with data mechanism techniques, the method performs image recognition and mathematical modeling on historical operating data of thermal and electrical loads of thermal power units. Image pixels are converted into time-series data with physical meaning. Based on first-order difference values, absolute thresholds, and stability criteria, long-term load curves are intelligently segmented into short line segments with single geometric features. Then, feature vectors are constructed from five dimensions: fitting slope, duration, linearity, root mean square error, and morphological fluctuation. Clustering algorithms are used to group line segments with similar features into the same operating condition category, thereby effectively distinguishing and identifying complex operating conditions of units under different combinations of electrical and thermal loads. This method provides a reliable basis for statistical analysis of unit operating conditions, comparative analysis of indicators, analysis of energy-saving potential, and safety analysis. It is applicable to pure condensing units and cogeneration units, and has strong versatility and innovation.

Description

Technical Field

[0001] This invention relates to the field of additive manufacturing technology, and in particular to a method for identifying the operating conditions of electrical loads in thermal power units. Background Technology

[0002] With a high proportion of renewable energy being integrated into the grid, thermal power units frequently participate in deep peak-shaving operations, resulting in electricity loads characterized by wide load ranges, significant fluctuations, and long periods of low-load operation. The National Development and Reform Commission and the National Energy Administration officially issued the "Implementation Plan for the Special Action to Upgrade New Generation Coal-fired Power (2025-2027)," outlining requirements for deep peak-shaving and load change rates for thermal power in response to the changing power landscape. This has created difficulties in analyzing and statistically analyzing the energy-saving potential of thermal power units. This invention utilizes big data technology to identify the operating conditions of thermal power units, providing a comparative basis for further statistical analysis of unit data under the same operating conditions, comparative analysis of indicators, analysis of energy-saving potential, and safety analysis. It possesses strong versatility and innovation. Summary of the Invention

[0003] The first aspect of this disclosure provides a method for identifying the operating conditions of electrical loads in thermal power units, comprising the following steps: S1: Extract electrical load time-series data from the historical load trend map of thermal power units, convert image pixels into time-series data, and filter and denoise the electrical load time-series data; S2: Based on the first-order difference value of load time series data, the preset absolute threshold of rate of change, and the stability criterion of continuous sampling points, the long-term electrical load time series curve is divided into short line segments with a single geometric feature. S3: Construct a multi-dimensional feature vector for each short line segment. The feature vector shall include at least five dimensions: fitting slope, duration, linearity, root mean square error, and morphological fluctuation. S4: Use clustering algorithms to cluster short line segments and group line segments with similar characteristics into the same working condition; S5: Based on the mean value of the characteristic parameters of each cluster in the clustering results, determine the operating condition type and output the identified operating condition of the thermal power unit.

[0004] In conjunction with the first aspect, S1 specifically includes: Set upper and lower limits for power generation load and a fixed time span; Historical data is retrieved from the thermal power control system or thermal power information monitoring system, where the X-axis curve represents the time range and the power generation load curve is marked in red. Extracting the red area using the HSV color space; The extracted red curve is subjected to skeleton extraction, and the curve is refined to a single pixel width; The extracted load sequence is filtered by moving average to eliminate high-frequency noise.

[0005] In conjunction with the first aspect, S2 uses the Ramer-Douglas-Peucker algorithm, sets a distance threshold ε, simplifies the electrical load time-series curve into a broken line, obtains a set of line segments, and removes invalid line segments that are too short in duration or have too small a power change.

[0006] In conjunction with the first aspect, in the multidimensional feature vector: the fitting slope is the slope of the best-fitting straight line obtained by performing linear regression on all points within the line segment using the least squares method, which characterizes the average rate of power change; The duration is the time difference between the end point and the beginning point of the line segment; Linearity R² is based on the formula

[0007] Calculation, where This is the actual power reading. To fit the theoretical power value on the straight line, This represents the average power. Root mean square error is based on the formula calculate; The morphological fluctuation is based on the formula The calculation involves dividing the current line segment into equal parts on the time axis. For each segment, calculate the local slope. , This represents the average local slope.

[0008] In conjunction with the first aspect, S4 employs the DBSCAN algorithm to find high-density point clusters in the three-dimensional feature space composed of fitting slope, linearity, and morphological fluctuations, and marks points far from any cluster as noise points.

[0009] In conjunction with the first aspect, S5 specifically includes: Calculate the characteristic mean of all line segments within each cluster; If the mean linearity of a cluster is greater than 0.95 and the mean morphological fluctuation is less than 0.1, it is marked as a standard linear process. If the mean linearity of a cluster is less than 0.8, it is marked as a nonlinear fluctuation process.

[0010] In conjunction with the first aspect, for heating units, based on the identification of electrical load operating conditions, a comprehensive weighted average dimension of the impact of heating on coal consumption is added.

[0011] In conjunction with the first aspect, the duration is set to be greater than 1 hour.

[0012] This disclosure provides a second aspect of improving an electronic device, comprising: One or more processors; A storage unit is used to store one or more programs, which, when executed by one or more processors, enable the one or more processors to implement the method for identifying the operating conditions of thermal power unit electrical load.

[0013] A third aspect of this disclosure provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, enables the method for identifying operating conditions based on the electrical load of a thermal power unit.

[0014] Beneficial Effects: This disclosure provides a method for identifying the operating conditions of thermal power units based on electrical load. By combining image contour extraction with data mechanism techniques, it performs image recognition and mathematical modeling on historical operating data of thermal and electrical loads of thermal power units. The image pixels are converted into time-series data with physical meaning. Based on first-order difference values, absolute thresholds, and stability criteria, long-term load curves are intelligently segmented into short line segments with single geometric features. Then, feature vectors are constructed from five dimensions: fitting slope, duration, linearity, root mean square error, and morphological fluctuation. Clustering algorithms are used to group line segments with similar features into the same operating condition category, thereby effectively distinguishing and identifying complex operating conditions of units under different combinations of electrical and thermal loads. This achieves the technical effect of providing a reliable comparative basis for statistical analysis of unit operating conditions, comparative analysis of indicators, analysis of energy-saving potential, and safety analysis. It is applicable to pure condensing units and cogeneration units, and has strong versatility and innovation. Attached Figure Description

[0015] Figure 1 This is a schematic flowchart of a method for evaluating the hot cracking sensitivity of L-PBF alloy with high crack sensitivity according to an embodiment of the present disclosure. Figure 2 An electronic device according to an embodiment of this disclosure. Detailed Implementation

[0016] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those disclosed herein.

[0017] The terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the present disclosure. The singular forms “a,” “the,” and “the” as used in this disclosure and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0018] Figure 1 This is a flowchart illustrating a method for identifying the operating conditions of electrical loads in thermal power units according to an embodiment of this disclosure, including the following steps: S1: Extract electrical load time-series data from the historical load trend map of thermal power units, convert image pixels into time-series data, and filter and denoise the electrical load time-series data; S1 specifically includes: Set upper and lower limits for power generation load and a fixed time span; Historical data is retrieved from the thermal power control system or thermal power information monitoring system, where the X-axis curve represents the time range and the power generation load curve is marked in red. Extracting the red area using the HSV color space; The extracted red curve is subjected to skeleton extraction, and the curve is refined to a single pixel width; The extracted load sequence is filtered by moving average to eliminate high-frequency noise.

[0019] In this embodiment, the upper and lower limits of the power generation load of the thermal power unit and a fixed time span are first set. This setting is used for coordinate alignment and range clipping during subsequent image recognition, so that the load curves of different time periods have a unified coordinate reference, which facilitates consistency when converting image pixels into physical quantities.

[0020] Retrieve historical load trend charts from the distributed control system (DCS) or plant-level monitoring information system (SIS) of the thermal power unit. In this trend chart, the horizontal axis (X-axis) represents the time range, and the vertical axis represents the power generation load value. The power generation load curve is specifically drawn with a red line to distinguish it from other parameter curves (such as main steam pressure, temperature, etc.).

[0021] The historical trend chart above is processed using the HSV color space model. A threshold range for the red hue is set, and all pixels belonging to the red region are extracted, thus separating the power generation load curve from the complex background and multi-colored curves. This step effectively eliminates interference from non-red curves, retaining only the pixel set corresponding to the target load curve.

[0022] The extracted red curve pixel region undergoes skeleton extraction processing. The skeleton extraction algorithm progressively thins the red curve, which has a certain width, until the curve width is reduced to a single pixel. This processing eliminates the ambiguity caused by the original curve being too thick, which might correspond to multiple vertical coordinates at the same time point, ensuring that each time point corresponds to only one unique load pixel position.

[0023] The refined single-pixel curves are transformed into time-series data with physical meaning. Specifically, the horizontal coordinate (corresponding to time) of each pixel is traversed along the X-axis, and the vertical coordinate (corresponding to the number of pixel rows) of that pixel is read. Based on the pre-set upper and lower limits of load and coordinate mapping relationship, the vertical coordinate pixel value is converted into the actual power generation load value, thereby obtaining a high-precision (time, load) coordinate point set.

[0024] The extracted load sequence is then subjected to a moving average filter. The size of the filter window is set (e.g., 5 to 11 sampling points, the specific value depending on the sampling frequency and noise level of the original data). The arithmetic mean of the load values ​​within each window is calculated, and this average value replaces the original load value at the center point of the window. Moving average filtering effectively eliminates high-frequency random noise introduced during sensor acquisition while maintaining the overall trend of the load curve, thus obtaining smooth and reliable time-series load data, providing a high-quality data foundation for subsequent intelligent segmentation processing.

[0025] S2: Based on the first-order difference value of load time series data, the preset absolute threshold of rate of change, and the stability criterion of continuous sampling points, the long-term electrical load time series curve is divided into short line segments with a single geometric feature. S2 uses the Ramer-Douglas-Peucker algorithm, sets a distance threshold ε, simplifies the electrical load time-series curve into a broken line, obtains a set of line segments, and removes invalid line segments that are too short in duration or have too small a power change.

[0026] In step S2, the obtained smoothed load time series data is first subjected to first-order difference calculation, that is, the load difference between two adjacent time points is calculated at each sampling point to obtain the load change rate sequence. A preset absolute threshold for the change rate is set, which is determined based on the percentage of the unit's rated power (e.g., 1% to 3% of the rated power) or actual engineering experience. When the absolute value of the first-order difference is lower than the threshold for multiple consecutive sampling points, the current time period is determined to be in a stable operating state; conversely, when the absolute value of the first-order difference exceeds the threshold, the load is determined to have entered a change phase.

[0027] Based on the aforementioned stability criteria, and combined with the changes in the first-order difference value, inflection points or state transition points in the load curve are identified. Specifically, when the load change rate changes from below a threshold to above a threshold, or from above a threshold to below a threshold, and this state change continues for more than a preset number of sampling points (e.g., 3 to 5 consecutive points), this location is marked as a candidate boundary point for the curve segment. In this way, the long-term load curve is initially divided into several candidate segments, each segment having relatively consistent load change characteristics.

[0028] Based on the initial segmentation, the Ramer-Douglas-Peucker (RDP) algorithm is further used to precisely compress and segment the load curve. The entire load time-series data set is used as input, and a distance threshold ε is set (this value can be set according to the percentage of the load range, for example, 0.5% to 2% of the rated load). The RDP algorithm recursively finds the point on the curve farthest from the line connecting the two endpoints. If this distance is greater than ε, the point is used as the new segmentation point, dividing the curve into two segments. This process is repeated for each segment until the distance from all segmentation points to the corresponding endpoints is less than ε. After processing by the RDP algorithm, the original curve is simplified into a series of polyline segments connected by key points, each polyline segment corresponding to a short line segment with a single geometric feature.

[0029] After obtaining the above set of line segments, the segments need to be screened for validity. Segments with excessively short durations, such as those less than 1 minute or less than the preset minimum effective duration (which can be set according to the unit's response characteristics), should be removed. Segments with excessively small power changes should also be removed, such as those where the difference between the start and end loads is less than 0.5% of the rated load or less than the preset minimum absolute change. These invalid segments typically correspond to measurement noise, minor fluctuations, or non-substantial load adjustments and should not be considered as independent operating conditions in subsequent feature extraction and cluster analysis. Each segment in the final retained set has a clear start and end time, start and end load values, and significant geometric morphological characteristics, providing accurate input data for the multi-dimensional feature engineering in step S3.

[0030] S3: Construct a multi-dimensional feature vector for each short line segment. The feature vector shall include at least five dimensions: fitting slope, duration, linearity, root mean square error, and morphological fluctuation. In the multidimensional feature vector: the fitting slope is the slope of the best-fitting straight line obtained by linear regression of all points within the line segment using the least squares method, which characterizes the average rate of power change; The duration is the time difference between the end point and the beginning point of the line segment; Linearity R² is based on the formula

[0031] Calculation, where This is the actual power reading. To fit the theoretical power value on the straight line, This represents the average power. Root mean square error is based on the formula calculate; The morphological fluctuation is based on the formula The calculation involves dividing the current line segment into equal parts on the time axis. For each segment, calculate the local slope. , This represents the average local slope.

[0032] The duration is set to be greater than 1 hour.

[0033] In step S3, for each valid short line segment obtained in step S2, a five-dimensional feature vector is constructed to quantitatively describe the geometric shape and variation characteristics of the electrical load curve segment corresponding to that line segment. The five dimensions are the fitting slope, duration, linearity, root mean square error, and morphological fluctuation.

[0034] To calculate the fitted slope, a least squares method is used to perform linear regression on all sampling points within the line segment. Let the line segment contain m sampling points, and the time coordinates of each point be... The corresponding load value is (i=1,2,…,m). Based on the least squares principle, the best-fitting straight line y=k·t+b is found to minimize the sum of squared residuals. The slope k represents the average rate of change of electrical load along this line segment, expressed in megawatts per minute (MW / min). This slope reflects the overall trend of load increase or decrease during this period: a positive k indicates a load increase, a negative k indicates a load decrease, and k close to zero indicates a relatively stable load.

[0035] For the calculation of duration, the time difference between the end and start of the line segment is directly taken. Considering the actual operating characteristics of thermal power units, this embodiment sets the duration to be greater than 1 hour before including it in the subsequent analysis. Line segments with too short a duration (e.g., less than 1 hour) may correspond to brief load disturbances or rapid adjustment processes, and their statistical significance and energy-saving analysis value are limited, so they are excluded or marked separately.

[0036] For linearity R², according to the formula... , Calculation, where This is the actual power reading. To fit the theoretical power value on the straight line, This represents the average power. The value of is between 0 and 1. The closer the value is to 1, the more closely the actual load points are distributed around the fitted line, and the higher the linearity of the line segment; conversely, The value of is less than 1. The smaller the value, the stronger the fluctuation of the load curve during that period, and the greater the deviation from a straight line.

[0037] The following formula is used to calculate the root mean square error (RMSE): , RMSE, expressed in megawatts (the same unit as the original load), characterizes the average vertical deviation between the actual data points and the fitted straight line. A smaller RMSE indicates a smoother load curve that closely follows the ideal straight line; a larger RMSE indicates that while the overall trend of the curve may be linear, there are significant local fluctuations or periodic oscillations.

[0038] For morphological fluctuations The calculation involves dividing the current line segment into n equal segments on the time axis (n can be flexibly set according to the total duration of the line segment; for example, n=4 for a duration of 1 hour, and n=6 for a duration of 2 hours, to ensure that each segment has a sufficient number of sampling points). The local slope of each segment is calculated (using the least squares method), and then the average of these n local slopes is calculated. Finally, calculate the standard deviation: , It reflects the degree of dispersion of the slope change within the line segment. A value close to 0 indicates that the slopes of the smaller segments in the entire line segment are almost the same, and the load changes at a uniform rate, which is a true straight line shape. The larger value indicates that although the overall fitted line is relatively large... The curve may appear higher, but it is actually a "pseudo-straight line" (such as an S-shaped curve or a stepped curve) composed of multiple short broken lines with different slopes. This microscopic morphological difference is achieved through... It can be effectively captured.

[0039] After completing the calculations for the above five dimensions, each effective short line segment is mapped to a five-dimensional feature vector. These feature vectors will serve as the input for the high-level clustering in step S4, which will be used to classify line segments with the same or similar shapes into the same operating condition.

[0040] S4: Use clustering algorithms to cluster short line segments and group line segments with similar characteristics into the same working condition; S4 uses the DBSCAN algorithm to find high-density point clusters in the three-dimensional feature space composed of fitting slope, linearity, and morphological fluctuations, and marks points far away from any cluster as noise points.

[0041] In step S4, for each short line segment's five-dimensional feature vector generated in step S3, the three most discriminative dimensions—fitting slope, linearity, and morphological fluctuation—are selected to construct a three-dimensional feature space. This selection is based on the following considerations: duration is mainly used to filter effective line segments; root mean square error (RMSE) is correlated with linearity and morphological fluctuation; and to avoid dimensional redundancy interfering with clustering results, the three core dimensions with stronger physical orthogonality are prioritized.

[0042] The density-based spatial clustering algorithm DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is used to perform unsupervised clustering of sample points in three-dimensional space. The DBSCAN algorithm does not require pre-specifying the number of classes, can automatically discover clusters of arbitrary shapes, and has the ability to identify noise points. These characteristics are highly compatible with the diversity and uncertainty of the load conditions of thermal power units.

[0043] Before performing DBSCAN clustering, the feature values ​​of the three dimensions need to be normalized. Because the magnitude of the fitted slope (typically several megawatts per second or minute) differs significantly from linearity (between 0 and 1) and morphological fluctuation (typically less than 1), directly calculating the Euclidean distance would cause the fitted slope dimension to excessively dominate the clustering results. Therefore, the Min-Max normalization method is used to map the value range of each dimension to the [0,1] interval, ensuring that the three dimensions have equal weight in the distance calculation.

[0044] Two key parameters for the DBSCAN algorithm are defined: neighborhood radius (eps) and minimum number of samples (minPts). The neighborhood radius eps defines the adjacency range of a sample point; a point is considered a core point if it contains a sufficient number of other sample points within this range. Considering the actual operating characteristics of thermal power units, eps can be set to a value between 0.1 and 0.3, with the specific value determined using a K-distance graph based on the density distribution in the feature space. The minimum number of samples (minPts) is generally set to 3 to 5, indicating that a core point must contain at least 3 samples, including itself, to ensure the statistical significance of the clustering.

[0045] The algorithm executes as follows: It traverses all sample points in the 3D feature space (each sample corresponds to a short line segment). For unvisited points, it calculates the number of sample points contained in the neighborhood of its eps. If this number is greater than or equal to minPts, the point is marked as a core point, and the algorithm expands outward from this point, grouping all density-reachable sample points into the same cluster. If the number of sample points in the neighborhood is less than minPts but the point is located in the neighborhood of a core point, it is marked as a boundary point and grouped into the corresponding cluster. If the point is neither a core point nor a neighborhood of any core point, it is marked as a noise point. This process is iterated until all sample points have been visited.

[0046] After clustering, the following results are output: Each identified cluster represents a type of electrical load operating condition with similar geometric features; samples marked as noise points correspond to load curve segments with rare morphologies, extremely low frequency of occurrence, or abnormal data, such as distorted curves caused by measurement faults, or large fluctuations in a very short time. Noise points are not included in subsequent operating condition statistics and comparative analysis, but can be used alone for anomaly detection or data quality auditing.

[0047] It is worth noting that DBSCAN searches for high-density point groups in a three-dimensional space composed of fitting slope, linearity, and morphological fluctuation, effectively overcoming the limitations of single-index judgment. For example, based solely on two line segments with similar fitting slopes, one might have extremely high linearity and extremely low morphological fluctuation (perfect straight-line climbing), while the other has lower linearity and greater morphological fluctuation (oscillating climbing). The distance between the two in three-dimensional space will naturally increase, thus classifying them into different clusters and achieving a fine-grained distinction of operating conditions.

[0048] S5: Based on the mean value of the characteristic parameters of each cluster in the clustering results, determine the operating condition type and output the identified operating condition of the thermal power unit.

[0049] S5 specifically includes: Calculate the characteristic mean of all line segments within each cluster; If the mean linearity of a cluster is greater than 0.95 and the mean morphological fluctuation is less than 0.1, it is marked as a standard linear process. If the mean linearity of a cluster is less than 0.8, it is marked as a nonlinear fluctuation process.

[0050] In step S5, post-processing analysis is performed on each cluster obtained from the clustering in step S4, calculating the statistical mean of all line segments within the cluster across each feature dimension. Specifically, for each cluster, the average fitted slope, average duration, average linearity, average root mean square error, and average morphological fluctuation of all short line segments contained therein are calculated. These means reflect the comprehensive characteristics of the typical operating conditions represented by the cluster.

[0051] Based on the above mean calculation, a preset logical rule is used to label the working condition type of each cluster. This embodiment sets two main discrimination criteria: The first category of criteria targets standard operating processes with excellent linear characteristics. When the mean linearity of a cluster is greater than 0.95 and the mean morphological fluctuation is less than 0.1, the cluster is marked as a "standard linear process." This label means that the electrical load curve segment in this cluster has extremely high linear fit, the actual load points are closely distributed around the fitted straight line, and the slope of each internal segment changes very little. The load rises or falls at a uniform and stable rate (or remains stable near a certain load level). This type of operating condition corresponds to the typical smooth load change process or stable operation process of thermal power units and is an ideal benchmark operating condition for energy-saving analysis and index comparison.

[0052] The second category of criteria targets fluctuating processes with significant nonlinear characteristics. When the mean linearity of a cluster is less than 0.8, the cluster is labeled as a "nonlinear fluctuating process." This label means that the overall linearity of the electrical load curve segment in this cluster is poor, and the actual load point deviates significantly from the fitted straight line, reflecting irregular load changes, oscillations, or external disturbances during that period. This type of operating condition can be further subdivided into large fluctuations during deep peak shaving, step-like changes caused by frequent changes in load commands, or unstable operating conditions, which can serve as analytical clues for equipment performance degradation or abnormal control parameters.

[0053] For clusters that do not meet the above two criteria (i.e., the mean linearity is between 0.8 and 0.95, or the linearity is greater than 0.95 but the morphological fluctuation is greater than or equal to 0.1), intermediate types can be marked according to actual needs, such as "general linear process" or "quasi-linear fluctuating process," or they can be reviewed and labeled by human experts. These intermediate types usually correspond to working conditions where the load change has a certain linear trend but is not ideal, or exhibits a combination of S-shaped, stepped, or other morphological forms.

[0054] After determining the operating condition type, the identification results are output. Output formats include, but are not limited to: generating a label for each operating condition and its corresponding typical characteristic parameters; labeling each time period in the original load curve with the corresponding operating condition type according to time sequence, forming an operating condition time-series distribution map; and summarizing the data segments of all line segments under the same operating condition for subsequent statistical analysis of data under the same operating condition, comparative analysis of indicators, energy-saving potential analysis, and safety analysis. For heating units, the output results can also include comprehensive weighted average information related to heating, achieving multi-dimensional operating condition classification under combined heat and power conditions.

[0055] Thus, through five steps—data extraction and cleaning, intelligent segmentation, multi-dimensional feature engineering, high-level clustering, and result verification and output—automatic identification of operating conditions based on the electrical load of thermal power units has been fully realized. This method does not rely on manual experience to set fixed load ranges, but automatically discovers operating conditions with recurring characteristics and specific geometric patterns from historical data, exhibiting strong adaptability and generalization capabilities.

[0056] For heating units, in addition to identifying the electrical load operating conditions, a comprehensive weighted average dimension of the impact of heating on coal consumption is added.

[0057] In practice, historical data related to heating is first obtained from the thermal power plant control system or plant-level monitoring information system. This data includes, but is not limited to, heating steam extraction flow rate (unit: tons / hour), extraction steam pressure, and extraction steam temperature, or the heating heat (unit: gigajoules / hour) can be directly obtained. Considering that heating parameters usually change relatively smoothly and their impact on coal quantity can be approximated as a linear or quasi-linear relationship, this embodiment adopts an average approximation method. That is, within the time interval corresponding to each identified short segment of electrical load, the arithmetic mean of heating-related parameters is calculated.

[0058] Then, the comprehensive weight of the heating parameters on the coal quantity is calculated. One of two equivalent methods can be used: The first method is to establish an empirical conversion factor between the heating extraction steam flow rate (or heat) and the standard coal consumption. For example, based on historical performance test data or design parameters of the unit, determine the standard coal consumption increment (unit: grams / ton) corresponding to each ton of heating extraction steam, or the standard coal quantity (unit: grams / gigajoule) corresponding to each gigajoule of heating heat. Multiply this conversion factor by the average heating parameters over that period to obtain the equivalent coal quantity increment caused by heating. The second method is to directly use a normalized heating load intensity index, for example, setting the maximum heating extraction steam flow rate to 1, and using the ratio of the actual extraction steam flow rate to the maximum value as the weight value, ranging from 0 to 1.

[0059] The average weighted average of the calculated impact of heating on coal consumption is denoted as... The average weight can be used as an independent dimension, appended to the five-dimensional feature vector constructed in step S3, to form a six-dimensional feature vector. Alternatively, a two-stage clustering strategy can be adopted: first, perform a first clustering based on the five-dimensional feature vector of electrical load to obtain several electrical load condition clusters; then, within each electrical load condition cluster, perform a second sub-clustering based on the average heating weight (e.g., using one-dimensional DBSCAN or setting a threshold interval), thereby distinguishing different heating load levels under the same electrical load change pattern.

[0060] For example, given two operating periods with the same characteristics of "load increase rate of approximately 3 MW / min, duration of 2 hours, and linearity of 0.98," if the average steam extraction flow rate for heating is 50 tons / hour in one period and 150 tons / hour in the other, then due to the different impacts of heating on coal consumption, the boiler combustion state, turbine flow characteristics, and overall efficiency will differ significantly between the two periods, and they should not be considered identical operating conditions. By adding a heating weight dimension, the two periods will be classified into different subcategories, thus providing a more accurate benchmark for subsequent energy-saving potential analysis and comparison with similar operating conditions.

[0061] For units with large or frequent fluctuations in heating parameters, the standard deviation or fluctuation range of heating parameters during that period can be added to the calculation of the comprehensive weighted average to characterize the stability of the heating load. This information can be used as an optional auxiliary dimension to further improve the accuracy of operating condition classification.

[0062] In summary, this method expands the original pure electric load condition identification system into a combined electric and heat condition identification system applicable to cogeneration units by adding a comprehensive weighted average dimension of the impact of heating on coal quantity. Without changing the core algorithm framework, it achieves full coverage of all types of thermal power units, including pure condensing units and heating units, and has broad adaptability and engineering practical value.

[0063] Electronic device 200 can be a desktop computer, laptop, handheld computer, cloud server, or other electronic device. Electronic device 200 may include, but is not limited to, processor 201 and memory 202. Those skilled in the art will understand that... Figure 2 This is merely an example of electronic device 200 and does not constitute a limitation on electronic device 200. It may include more or fewer components than shown, or combine certain components, or different components. For example, electronic device may also include input / output devices, network access devices, buses, etc.

[0064] The processor 201 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0065] The memory 202 can be an internal storage unit of the electronic device 200, such as a hard disk or RAM of the electronic device 200. The memory 202 can also be an external storage device of the electronic device 200, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the electronic device 200. Furthermore, the memory 202 can include both internal and external storage units of the electronic device 200. The memory 202 is used to store the computer program 203 and other programs and data required by the electronic device. The memory 202 can also be used to temporarily store data that has been output or will be output.

[0066] The above embodiments are only used to illustrate the technical solutions of this disclosure, and are not intended to limit it. Although this disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this disclosure, and should all be included within the protection scope of this disclosure.

Claims

1. A method for identifying the operating conditions of electrical load in thermal power units, characterized in that, Includes the following steps: S1: Extract electrical load time-series data from the historical load trend map of thermal power units, convert image pixels into time-series data, and filter and denoise the electrical load time-series data; S2: Based on the first-order difference value of load time series data, the preset absolute threshold of rate of change, and the stability criterion of continuous sampling points, the long-term electrical load time series curve is divided into short line segments with a single geometric feature. S3: Construct a multi-dimensional feature vector for each short line segment. The feature vector shall include at least five dimensions: fitting slope, duration, linearity, root mean square error, and morphological fluctuation. S4: Use clustering algorithms to cluster short line segments and group line segments with similar characteristics into the same working condition; S5: Based on the mean value of the characteristic parameters of each cluster in the clustering results, determine the operating condition type and output the identified operating condition of the thermal power unit.

2. The method according to claim 1, characterized in that, S1 specifically includes: Set upper and lower limits for power generation load and a fixed time span; Historical data is retrieved from the thermal power control system or thermal power information monitoring system, where the X-axis curve represents the time range and the power generation load curve is marked in red. Extracting the red area using the HSV color space; The extracted red curve is subjected to skeleton extraction, and the curve is refined to a single pixel width; The extracted load sequence is filtered by moving average to eliminate high-frequency noise.

3. The method according to claim 1, characterized in that, S2 uses the Ramer-Douglas-Peucker algorithm, sets a distance threshold ε, simplifies the electrical load time-series curve into a broken line, obtains a set of line segments, and removes invalid line segments that are too short in duration or have too small a power change.

4. The method according to claim 1, characterized in that, In the multidimensional feature vector: the fitting slope is the slope of the best-fitting straight line obtained by linear regression of all points within the line segment using the least squares method, which characterizes the average rate of power change; The duration is the time difference between the end point and the beginning point of the line segment; Linearity R² is based on the formula ， Calculation, where This is the actual power reading. To fit the theoretical power value on the straight line, This represents the average power. Root mean square error is based on the formula calculate; The morphological fluctuation is based on the formula The calculation involves dividing the current line segment into equal parts on the time axis. For each segment, calculate the local slope. , This represents the average local slope.

5. The method according to claim 1, characterized in that, S4 uses the DBSCAN algorithm to find high-density point clusters in the three-dimensional feature space composed of fitting slope, linearity, and morphological fluctuations, and marks points far away from any cluster as noise points.

6. The method according to claim 1, characterized in that, S5 specifically includes: Calculate the characteristic mean of all line segments within each cluster; If the mean linearity of a cluster is greater than 0.95 and the mean morphological fluctuation is less than 0.1, it is marked as a standard linear process. If the mean linearity of a cluster is less than 0.8, it is marked as a nonlinear fluctuation process.

7. The method according to claim 1, characterized in that, For heating units, in addition to identifying the electrical load operating conditions, a comprehensive weighted average dimension of the impact of heating on coal consumption is added.

8. The method according to claim 1, characterized in that, The duration is set to be greater than 1 hour.

9. An electronic device, characterized in that, include: One or more processors; A storage unit for storing one or more programs that, when executed by one or more processors, enable the one or more processors to implement the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it can implement the method according to any one of claims 1 to 8.

Images