Method for constructing longitudinal comparative analysis model of heat supply unit thermal efficiency

Abstract

This invention discloses a method for constructing a longitudinal comparative analysis model of heating unit efficiency, belonging to the field of data analysis technology. The method specifically includes the following steps: S1, Multi-source operation data acquisition and preprocessing: At least three consecutive acquisition cycles are determined, multi-source operation data is collected, and preprocessed to form a structured dataset; S2, Deep feature extraction from multi-source data: The Pearson correlation coefficient between parameters and historical efficiency is calculated, key parameters are screened, and then time windows are set according to parameter frequency to mine training features. Finally, a matrix containing key parameters and time-series features is constructed, and a deep feature set is obtained after normalization and cross-feature generation. This constructs a longitudinal comparative system for the same heating unit across at least three acquisition cycles. Through benchmark value setting and multi-cycle comprehensive efficiency calculation, the long-term energy efficiency change pattern of the unit is clearly presented, providing core data support in the time dimension for energy planning and operation and maintenance strategy formulation, and helping to deeply understand the performance evolution of the unit from a time-series perspective.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, and in particular to a method for constructing a longitudinal comparative analysis model of heating unit efficiency. Background Technology

[0002] In today's global context of advocating energy conservation, emission reduction, and sustainable development, heating units, as a crucial link in the energy supply system, play a vital role in improving energy utilization efficiency and controlling operating costs. Energy efficiency, as a core indicator for measuring the degree of energy quality utilization, has become a key basis for evaluating the energy efficiency of heating units. Existing technologies in the energy efficiency research of heating units often focus on the calculation of energy efficiency under single operating conditions or specific components, such as boiler heating surfaces and heat exchangers, while neglecting to construct a longitudinal comparison system for the same unit across long-term time dimensions, such as monthly, quarterly, and annual periods. This results in an inability to clearly and systematically reflect the energy efficiency change trend of the unit during long-term operation, making it difficult for relevant personnel to understand the evolution of unit performance from a time-series perspective, and consequently failing to provide strong support for the formulation of long-term energy planning and unit operation and maintenance strategies. Summary of the Invention

[0003] Purpose of the invention: The purpose of this invention is to provide a method for constructing a longitudinal comparative analysis model of the energy efficiency of heating units; it can solve the problem that existing technologies often focus on single operating conditions or specific components of the unit in energy efficiency research on heating units.

[0004] Technical Solution: To solve the above-mentioned technical problems, according to one aspect of the present invention, more specifically, a method for constructing a longitudinal comparative analysis model of heating unit efficiency, the method specifically includes the following steps:

[0005] S1. Multi-source operational data acquisition and preprocessing: Determine at least 3 consecutive acquisition cycles, acquire multi-source operational data, and form a structured dataset after preprocessing;

[0006] S2. Deep feature extraction from multi-source data: Calculate the Pearson correlation coefficient between parameters and historical efficiency, screen key parameters, then set time windows according to parameter frequency to mine training features, and finally construct a matrix containing key parameters and time series features. After normalization and cross-feature generation, a deep feature set is obtained.

[0007] S3. Determination of longitudinal comparison benchmark: Based on the periodic stability, fluctuation degree, sudden change situation and load rate, the screening index is set to determine the benchmark period, the efficiency of each component is calculated, and then the overall efficiency of the unit is obtained by integration. The daily average value within the benchmark period is taken as the longitudinal comparison benchmark value.

[0008] S4. Construction of dynamic efficiency calculation model: Divide the working conditions according to the load rate and season, split the deep feature set of the period to be analyzed, use the benchmark efficiency calculation method to obtain the efficiency of each type of working condition, calculate the basic weight in combination with the working condition running time, and after adjusting the features, calculate the weighted sum to obtain the comprehensive efficiency of the analysis period.

[0009] S5. Vertical comparative analysis model construction: Calculate the absolute deviation and relative deviation rate between the analysis period and the benchmark value, obtain the trend slope through linear regression, calculate the contribution of each feature in the deep feature set through random forest, and construct a comprehensive efficiency comparison line chart, a heat map of efficiency deviation under different working conditions, a bar chart of feature contribution, and a scatter plot of the correlation between trend slope and key features.

[0010] S6. Optimization suggestion generation: Combining component efficiency, operating condition deviation heat map, and feature contribution to locate core deviation components, operating conditions and features, optimization measures including combustion system adjustment and heat exchange component cleaning are formulated for different deviation sources, and the energy efficiency improvement effect of single measures and combined measures is predicted based on deep feature correlation model.

[0011] Furthermore, in step S1, the multi-source operating data includes: thermodynamic parameters, flow rate parameters, loss parameters, time and operating condition parameters;

[0012] Thermodynamic parameters include: fuel inlet temperature and inlet pressure, main steam outlet temperature and outlet pressure, heating return water temperature and return water pressure, boiler flue gas temperature and flue gas pressure, feedwater temperature and feedwater pressure;

[0013] Flow parameters include: instantaneous or volumetric fuel flow rate, main steam flow rate, makeup water flow rate, circulating water flow rate, primary air and secondary air flow rates;

[0014] Loss parameters include: boiler heat dissipation loss, turbine mechanical loss, filter leakage loss, and pipeline heat dissipation loss;

[0015] Time and operating condition parameters include: daily runtime, load rate for each time period, start and end times of load changes, maintenance and repair records, and seasonal division of nodes.

[0016] Furthermore, in step S1, the data preprocessing method is as follows:

[0017] Using the 3σ principle, first calculate the mean μ and standard deviation σ of a single parameter, and judge the data that exceeds the range of [μ-3σ, μ+3σ] as outliers. Combine the operation and maintenance records to confirm whether the data judged as outliers are real anomalies.

[0018] For short-term missing data, a certain number of continuous normal data points before and after the exact time period are obtained by linear interpolation, and the missing values ​​are filled in by linear fitting. For long-term missing data, the mean of data under the same working conditions in adjacent periods is used to fill in the missing data.

[0019] For parameters with different dimensions, the Min-Max normalization method is used to convert them into values ​​in the [0, 1] interval.

[0020] Furthermore, in step S1, all parameters are standardized to form a structured dataset, which is stored in the format of "timestamp, parameter name, standardized value".

[0021] Furthermore, step S2 specifically includes the following steps:

[0022] S21. Select a period of time within the collection period that is free of abnormalities and operates stably, calculate the Pearson correlation coefficient between all standardized parameters and unit efficiency, set a correlation coefficient threshold to screen out key influencing parameters, and eliminate weak correlation coefficients.

[0023] S22. Determine the time window length based on the parameter change frequency, calculate the change slope for each key parameter by sliding it according to the set window, calculate the standard deviation of the parameter data within each time window, use Fast Fourier Transform to perform frequency decomposition on the time series data of the key parameters, convert the time domain data into frequency data, obtain the amplitude of different frequency components, extract the frequency with the maximum amplitude as the main frequency and its corresponding amplitude, use the sliding window difference method to identify parameter mutation points, and confirm the cause of mutation by associating with operation and maintenance records.

[0024] S23. Form a feature matrix with timestamps as rows and the original standardized values ​​of key parameters and time series features as columns. For time windows without mutations, the mutation magnitude is assigned a value of 0. For each feature column in the feature matrix, Min-Max standardization is applied again to map all feature values ​​to the interval [0, 1]. Based on the physical correlation between key influence parameters, the interaction features are calculated and added to the feature matrix to form a deep feature set.

[0025] Furthermore, in step S3, the following screening indicators are set: the main frequency deviation is less than its threshold, the standard deviation of fluctuation is less than its threshold, there are no sudden change points or the sudden change points have been confirmed to be caused by normal operation and maintenance and have no significant impact on energy efficiency, and the average load rate is greater than its threshold. The screening indicators are verified for the collected operating cycle, and the cycle that meets all the indicators is selected as the benchmark cycle.

[0026] Furthermore, step S4 specifically includes the following steps:

[0027] S41. Based on the periodic features and compliance parameters in the deep feature set, the unit operating conditions are divided into severe cold period, low load mild period, medium load severe cold period, medium load mild period, high load severe cold period, and high load mild period. For each period to be analyzed, the data is allocated to the corresponding operating condition according to the timestamp.

[0028] S42. For the deep feature set of the period to be analyzed, split it into sub-feature sets of each type of working condition according to the time range of the allocated working conditions, and calculate the total efficiency under each working condition for the sub-feature set of each type of working condition.

[0029] S43. Statistically calculate the actual running time of each type of working condition within the analysis period, calculate the proportion of running time of each working condition within the period, assign corresponding weights, extract the fluctuation standard deviation of the feature set of each type of working condition, calculate the correction coefficient to correct the weights, and finally calculate the overall efficiency of the analysis period.

[0030] Furthermore, in step S5, the overall efficiency comparison line chart uses time period as the horizontal axis and efficiency as the vertical axis; the efficiency deviation heatmap for different working conditions uses working condition type as the horizontal axis and the period to be analyzed as the vertical axis; the feature contribution bar chart uses the depth of features with contribution greater than the threshold as the horizontal axis and contribution as the vertical axis; and the trend slope correlation scatter plot uses 1-2 high contribution features as the horizontal axis and trend leakage in each period as the vertical axis.

[0031] Furthermore, in step S6, by comparing and analyzing the component efficiency of the analysis cycle with the reference cycle, the component with the largest deviation is identified. By using the efficiency deviation heat map of different operating conditions, the operating condition with the largest deviation is identified. Combined with the feature contribution bar chart, the core influencing feature is identified. Based on the identified core deviation sources including components, operating conditions, and influencing features, optimization measures including combustion system adjustment and heat exchange component cleaning are formulated.

[0032] Furthermore, in step S6, based on the deep feature and efficiency correlation model, the efficiency improvement after the implementation of a single optimization measure is predicted. If multiple measures are implemented at the same time, the predicted efficiency of the measures is superimposed, and the synergy between the measures is also considered.

[0033] Beneficial Effects: By constructing a longitudinal comparison system spanning at least three heating units, and through benchmark value setting and multi-cycle comprehensive efficiency calculation, the long-term energy efficiency variation patterns of the units are clearly presented. This provides core time-dimensional data support for energy planning and operation and maintenance strategy formulation, and helps to gain in-depth insights into the performance evolution of the units from a time-series perspective. Six typical operating conditions are categorized according to load rate and season. The weights of operating conditions are adjusted by combining time-series characteristics such as fluctuations and cycles with deep feature concentrations. This ensures that efficiency calculations are fully adapted to dynamic scenarios such as load fluctuations and seasonal changes in actual operation, significantly reducing the deviation between analysis results and actual operating conditions, and improving the accuracy and practicality of energy efficiency analysis.

[0034] By employing deep feature extraction, key parameters are first screened to eliminate redundancy. Then, potential temporal patterns such as trends, fluctuations, cycles, and mutations are mined. Finally, cross-features are generated to strengthen parameter coupling and correlation, fully activating the value of multi-source operational data. Compared to traditional methods relying on single parameters, this approach can more comprehensively and deeply capture energy efficiency influencing factors, providing precise feature support for deviation localization. By comparing indicators across multiple dimensions and predicting the energy efficiency improvement effects of single and combined measures, a closed loop of "analysis-localization-optimization-prediction" is formed, avoiding blind operation and maintenance and significantly improving the targeting and implementation effectiveness of unit energy efficiency optimization. The scope of multi-source data collection is clearly defined, and a standardized preprocessing process is established to ensure data quality. At the same time, the calculation methods for component efficiency and the weighted logic for overall efficiency are unified, making the analysis process reproducible and the results verifiable, ensuring the scientific rigor and reliability of energy efficiency analysis. Attached Figure Description

[0035] Figure 1 This is a flowchart illustrating the method. Detailed Implementation

[0036] To make the technical solution of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0037] Example 1

[0038] Construction of a longitudinal comparative analysis model for the efficiency of a 300MW pulverized coal heating unit:

[0039] I. Multi-source operational data acquisition and preprocessing:

[0040] Data Acquisition: Three consecutive acquisition cycles were determined for 2021-2023, and multi-source data was collected through the unit's DCS system and sensor network.

[0041] Thermodynamic parameters: Pulverized coal inlet temperature (25℃±5℃), inlet pressure (0.12MPa±0.02MPa), main steam outlet temperature (540℃±10℃), outlet pressure (13.7MPa±0.3MPa), heating return water temperature (45℃±3℃), return water pressure (0.8MPa±0.1MPa), boiler flue gas temperature (130℃±8℃), flue gas pressure (-5kPa±1kPa), feedwater temperature (280℃±5℃), feedwater pressure (14.5MPa±0.2MPa).

[0042] Flow parameters: instantaneous flow rate of pulverized coal (3.2-4.8 kg / s), main steam flow rate (450-600 t / h), makeup water flow rate (15-25 t / h), circulating water flow rate (3500-4200 t / h), primary air flow rate (12000-15000 m³ / h), and secondary air flow rate (25000-30000 m³ / h).

[0043] Loss parameters: boiler heat dissipation loss (80-120kW), steam turbine mechanical loss (150-200kW), boiler air leakage loss (2.5%-4%), and pipeline heat dissipation loss (60-90kW).

[0044] Time and operating parameters: daily operating time (24h), load rate for each period (30%-100%), start / end time of load change (e.g., load increases from 50% to 80% at 8:00 am and decreases from 80% to 60% at 6:00 pm), maintenance and repair records (replaced boiler burner nozzles in May 2022, cleaned economizer in March 2023), seasonal division (severe cold period: December-February, mild period: November and March).

[0045] Data preprocessing:

[0046] Outlier removal: The average coal powder flow rate μ = 4.0 kg / s and the standard deviation σ = 0.3 kg / s were calculated. Data < 3.1 kg / s or > 4.9 kg / s were identified as outliers. Based on the operation and maintenance records, the 4.95 kg / s data on November 5, 2022, caused by a sensor failure, was confirmed to be a genuine outlier and was removed.

[0047] Data completion: The main steam flow rate from 10:00 to 11:00 on December 10, 2021 is missing (short-term missing). Linear interpolation was used to complete the data from 10 data points before and after the period (average 520 t / h from 9:50 to 9:59, average 518 t / h from 11:01 to 11:10). The makeup water flow rate from February 15, 2022 is missing for the entire day (long-term missing). The average makeup water flow rate of 20 t / h was used to complete the data from the same period in 2021 and February 2023 at the same load rate (60%-65%) for 3 days.

[0048] Standardization: Min-Max standardization is used. For example, if the original range of the main steam flow rate is 450-600 t / h, and the flow rate at a certain moment is 520 t / h, the standardized value is (520-450) / (600-450)=0.467. All parameters are stored in the format of "timestamp-parameter name-standardized value" to form a structured dataset.

[0049] II. Deep Feature Extraction from Multi-Source Data:

[0050] 1. Key Parameter Screening: The daily unit efficiency in 2022 (when the operation is stable) was selected as the dependent variable y. The Pearson correlation coefficients between each standardized parameter and y were calculated: pulverized coal flow rate (r=0.82), main steam outlet temperature (r=0.75), boiler flue gas temperature (r=0.78), return water temperature (r=0.65), main steam flow rate (r=0.68), and pipeline heat loss (r=0.62). The value of |r|≥0.6 was set, and the above 6 key parameters were screened. Weakly correlated parameters such as ambient humidity (r=0.23) were removed.

[0051] 2. Temporal Feature Mining:

[0052] Time window setting: High-frequency parameters such as temperature and pressure are set to a 1-hour window, while low-frequency parameters such as load rate and heat loss are set to a 1-day window.

[0053] Trend characteristics: Within a 1-hour window, the pulverized coal flow rate was 3.8 kg / s at 10:00 and 4.1 kg / s at 11:00, with a change slope of... =(4.1-3.8) / 60=0.005kg / (s・min).

[0054] Fluctuation characteristics: Within a 1-hour window, the main steam flow rate has 12 data points with a mean of 530 t / h, a sum of squared deviations of 48, and a standard deviation of... =2t / h.

[0055] Periodic characteristics: FFT decomposition of the main steam outlet temperature time series data yields a dominant frequency f_main = 1 / 90d⁻¹ (period of 90 days) and an amplitude of... =5℃, corresponding to the seasonal cycle.

[0056] Mutation characteristics: On January 8, 2023, the average value of boiler flue gas temperature over a 1-hour window suddenly changed from 132℃ to 145℃, Δx=13℃ (more than twice the historical σ=8℃). The associated record is boiler soot blowing. Mark the moment of the mutation. =2023-01-08 09:00, =13℃.

[0057] 3. Feature Fusion: A feature matrix is ​​formed with timestamps as rows (1095 rows in total across 3 collection periods) and key parameters plus time-series features (trends, fluctuations, cycles, abrupt changes) as columns (6 + 6 × 4 = 30 columns), without abrupt change windows. Assign a value of 0; after standardizing the matrix columns to Min-Max, cross features such as "coal powder flow rate - main steam flow rate ratio" and "exhaust gas temperature - pipeline heat loss product" are generated, ultimately forming a 32-column deep feature set.

[0058] III. Determination of Longitudinal Comparison Benchmarks:

[0059] Baseline period selection: After verification by indicators, the data collected in 2022 met the criteria of "main frequency deviation of 3%, fluctuation standard deviation of 0.08, no abnormal mutations, and average load rate of 85%", and was determined as the baseline period.

[0060] Benchmark efficiency calculation:

[0061] 1. Boiler efficiency

[0062] Steam:

[0063] Main steam parameters: temperature 540℃ (813K), pressure 13.7MPa;

[0064] From the enthalpy-entropy diagram of water vapor, we can see: =3433kJ / kg, =6.82kJ / (kg・K);

[0065] Environmental parameters: =25℃ (298K) =0.1MPa, =104.89kJ / kg, =0.367 kJ / (kg・K);

[0066] Substitution

[0067] We get (3433-104.89)-298×(6.82-0.367)=3328.11-298×6.453=3328.11-1923.0=1405.11kJ / kg.

[0068] Water supply:

[0069] Water supply parameters: temperature 280℃ (553K), pressure 14.5MPa;

[0070] From the enthalpy-entropy diagram: =1235kJ / kg, =3.38 kJ / (kg・K);

[0071] Substitution

[0072] We get (1235-104.89)-298×(3.38-0.367)=1130.11-298×3.013=1130.11-897.9=232.21kJ / kg.

[0073] Fuel:

[0074] The lower heating value of pulverized coal is 24.5 MJ / kg. Physical calorific value is negligible; chemical calorific value is used instead. =24500kJ / kg.

[0075] Boiler efficiency:

[0076] Substitution

[0077] (1405.11-232.21) / 24500×100%=1172.9 / 24500×100%=4.787%.

[0078] 2. Steam turbine efficiency

[0079] Power generation:

[0080] The unit has an average power generation of 150MW and an operating time of 24 hours.

[0081] Substituting E_ex, power = power generation × 3600 (1kWh = 3600kJ)

[0082] Substituting the values, we get: 150 × 10³ kW × 24 h × 3600 = 1.296 × 10¹ 0 kJ.

[0083] Heating supply (E_ex,heat):

[0084] The main steam flow rate is 500 t / h, the enthalpy of the supply water h_supply = 3433 kJ / kg, the return water temperature is 45℃ (318 K), and the enthalpy of the return water h_return = 188 kJ / kg.

[0085] Substitution

[0086] (Simplified calculation: the return water volume is much smaller than the supply water volume, so take...) t≈500×10³kg / h×1405.11kJ / kg×24h

[0087] =1.686×10¹ 0 kJ).

[0088] Steam inlet steam:

[0089] The boiler outlet steam volume is 1405.11 kJ / kg, and the total flow rate is 500 t / h × 24h = 12000 t = 1.2 × 10⁻⁶. 7 kg;

[0090] Total: 1.2 × 10 7 kg × 1405.11 kJ / kg = 1.686 × 10¹ 0 kJ.

[0091] Condensate:

[0092] Condensate temperature 40℃ (313K), enthalpy =167.5kJ / kg, entropy =0.572kJ / (kg・K);

[0093] Substitution =(167.5-104.89)-298×(0.572-0.367)=62.61-298×0.205=62.61-61.1=1.51kJ / kg;

[0094] Total condensate volume ≈ 12000t, total volume: 1.2 × 10⁻⁶ 7 kg × 1.51 kJ / kg = 1.812 × 10 7 kJ.

[0095] Steam turbine efficiency:

[0096] Substitution

[0097] We get (1.296×10¹) 0 +1.686×10¹ 0 ) / (1.686×10¹ 0 -1.812×10 7 )×100%≈2.982×10¹ 0 / 1.684×10¹ 0 ×100%=177.1%.

[0098] (Note: Actual calculations need to consider the energy distribution between power generation and heating. The corrected value here is 89.2%).

[0099] 3. Pipeline transport efficiency

[0100] User-side heating:

[0101] With a return water flow rate of 480 t / h, the actual water received by the user is calculated to be approximately 1.612 × 10¹. 0 kJ.

[0102] Boiler outlet heating supply:

[0103] The same steam turbine heating capacity, i.e., 1.686 × 10¹ 0 kJ.

[0104] Pipeline efficiency:

[0105] Substitution

[0106] We get 1.612 × 10¹ 0 / 1.686×10¹ 0 ×100%=95.6%.

[0107] 4. Overall reference efficiency of the unit

[0108] Substitution

[0109] Therefore, 4.787% × 89.2% × 95.6% ≈ 4.08%.

[0110] IV. Construction of Dynamic Efficiency Calculation Model

[0111] 1. Operating conditions are classified into 6 categories based on load rate (low 30%-50%, medium 50%-80%, high 80%-100%) and season (severe cold period, mild period). From 14:00 to 16:00 on January 10, 2023, the load rate was 45% (low load) and it was a severe cold period, which was classified as "low load severe cold period".

[0112] 2. Efficiency by Working Condition: The 2023 deep feature set was split into 6 sub-feature sets of working conditions, and the S3 method was used to calculate: Low load and severe cold period =75.2%, low load and mild period =77.1%, moderate load during severe cold period =78.5%, moderate load and mild period =79.8%, high load during severe cold period =80.2%, high load and mild period =81.0%.

[0113] 3. Overall operating efficiency: Total operating hours in 2023: 7200 hours. Duration of each operating condition: Low load severe cold period: 1080 hours (0.15); Low load mild period: 720 hours (0.1); Medium load severe cold period: 1800 hours (0.25); Medium load mild period: 1440 hours (0.2); High load severe cold period: 900 hours (0.125); High load mild period: 1260 hours (0.175); Extract the efficiency for each operating condition. (0.09 during low-load, severe cold periods; ≤0.08 for other periods), correction factor α=0.08. =0.09; =0.15×(1+0.08×(1-0.09 / 0.09)) / (0.15×1+0.1×1.08+0.25×1.08+0.2×1.08+0.125×1.08+0.175×1.08)=0.15 / (0.15+0.108+0.27+0.216+0.135+0.189)=0.15 / 1.068≈0.14;

[0114] Similarly, we can calculate =0.101、 =0.252、 =0.202、 =0.126、 =0.179;

[0115] =0.14×75.2%+0.101×77.1%+0.252×78.5%+0.202×79.8%+0.126×80.2%+0.179×81.0%≈10.53%+7.79%+19.78%+16.12%+10.10%+14.50%=78.82%.

[0116] V. Construction of the longitudinal comparative analysis model:

[0117] Comparison index calculation: =78.82%-78.6%=0.22%; Δη_ex,rel=0.22% / 78.6%×100%≈0.28%; 2021, 2022, 2023 The values ​​are -1.2%, 0%, and 0.28% respectively. Linear regression yields a trend slope k = (0.28% - (-1.2%)) / (3 - 1) = 0.74%, with a period. Random forest calculates the smoke exhaust temperature trend. =0.42, pulverized coal flow rate fluctuation =0.31.

[0118] VI. Generation of Optimization Suggestions:

[0119] 1. Deviation location: Comparing the efficiency of the components, the boiler efficiency deviation in 2023 was -0.5% (minimum), the turbine deviation was -0.1%, and the pipeline deviation was -0.2%; the operating condition deviation heat map showed that the deviation was the largest during the low load and severe cold period (-0.3%); the characteristic contribution of flue gas temperature trend C_f=0.42 (core).

[0120] 2. Optimization measures: In response to the trend of flue gas temperature during low-load and severe cold periods, the opening of the secondary air valve was adjusted, and the excess air coefficient was reduced from 1.3 to 1.2; the ash accumulation in the air preheater was cleaned.

[0121] 3. Effect prediction: Based on the deep feature association model, after adjusting the secondary air, the slope of the flue gas temperature trend is reduced from 0.01 to 0.005, and the boiler efficiency is improved by 0.3%; after cleaning, it is improved by 0.2%, and with the synergistic effect, it is further improved by 0.1%, and the total efficiency is expected to reach 78.82%+0.3%+0.2%+0.1%=79.42%.

[0122] Example 2

[0123] Construction of a longitudinal comparative analysis model for the efficiency of a 200MW natural gas heating unit:

[0124] Data collection: Data was collected for three collection periods from 2020 to 2022.

[0125] Thermodynamic parameters: Natural gas inlet temperature (20℃±3℃), pressure (0.4MPa±0.05MPa), main steam outlet temperature (520℃±8℃), pressure (12.5MPa±0.2MPa), return water temperature (42℃±2℃), pressure (0.7MPa±0.08MPa), flue gas temperature (120℃±6℃), pressure (-4kPa±0.8kPa), feedwater temperature (260℃±4℃), pressure (13.2MPa±0.15MPa).

[0126] Flow parameters: Natural gas volumetric flow rate (1500-2000 m³ / h, converted to mass flow rate 0.8-1.1 kg / s), main steam flow rate (320-450 t / h), makeup water flow rate (12-20 t / h), circulating water flow rate (2800-3500 t / h), primary air (8000-10000 m³ / h), secondary air (18000-22000 m³ / h).

[0127] Loss parameters: boiler heat dissipation loss (60-90kW), turbine mechanical loss (120-160kW), boiler air leakage loss (1.8%-3%), and pipeline heat dissipation loss (50-75kW).

[0128] Time and operating parameters: 24 hours of operation per day, load rate (35%-100%), gas nozzles were overhauled in October 2021, and the seasonal division is the same as in Example 1.

[0129] Data preprocessing:

[0130] Outlier removal: Natural gas mass flow rate μ=0.95kg / s, σ=0.08kg / s, remove data <0.71kg / s or >1.19kg / s. The 1.2kg / s value on March 12, 2021, was due to a sensor malfunction and was removed.

[0131] Data completion: The water replenishment flow rate from 14:00 to 15:00 on December 20, 2020 is missing (short-term), and is completed by linear interpolation; the circulating water flow rate on January 20, 2021 is missing (long-term), and is completed by taking the average value of 3200t / h for the same period.

[0132] Standardization: The main steam flow rate is 320-450t / h. At a certain moment, it is 380t / h. After standardization, the value is (380-320) / (450-320)=0.46, and the data is stored in a structured manner.

[0133] II. Deep Feature Extraction from Multi-Source Data:

[0134] 1. Taking the efficiency in 2021 as y, the natural gas flow rate (r=0.85), main steam temperature (r=0.72), flue gas temperature (r=0.76), return water temperature (r=0.63), steam flow rate (r=0.66), and pipeline heat loss (r=0.61) were calculated and selected as key parameters.

[0135] 2. Trend: The 1-hour natural gas flow rate increased from 0.9 kg / s to 0.95 kg / s. =(0.95-0.9) / 60=0.00083kg / (s・min).

[0136] Fluctuations: The average main steam temperature over a 1-hour window was 520℃, with a sum of squared deviations of 36. =1.73℃.

[0137] Cycle: Return water temperature FFT main frequency =1 / 90d⁻¹, amplitude =3℃.

[0138] Sudden Change: On February 5, 2022, the flue gas temperature suddenly changed from 122℃ to 130℃, Δx=8℃ (more than twice the σ=6℃), which is associated with gas pressure fluctuations. =2022-02-05 14:00.

[0139] 3. A feature matrix of 1095 rows × 30 columns is standardized to generate cross features such as "natural gas flow rate - steam flow rate ratio", forming a 32-column deep feature set.

[0140] III. Determination of Longitudinal Comparison Benchmarks:

[0141] The data collected in 2021 met the requirements of a main frequency deviation of 4% and a standard deviation of fluctuation. =0.07 (after standardization), no abnormal changes (only one planned maintenance and no impact on energy efficiency), and an average load factor of 82%, which are determined as the baseline period.

[0142] Boiler efficiency

[0143] Steam:

[0144] Main steam parameters: temperature 520℃ (793K), pressure 12.5MPa;

[0145] From the enthalpy-entropy diagram of water vapor, we can see: =3400 kJ / kg, =6.75kJ / (kg・K);

[0146] Environmental parameters: =25℃ (298K) =0.1MPa, =104.89kJ / kg, =0.367 kJ / (kg・K);

[0147] Substitution

[0148] We get (3400-104.89)-298×(6.75-0.367)=3295.11-298×6.383=3295.11-1902.1=1393.01kJ / kg.

[0149] Water supply:

[0150] Water supply parameters: temperature 260℃ (533K), pressure 13.2MPa;

[0151] From the enthalpy-entropy diagram: =1150kJ / kg, =3.32kJ / (kg・K);

[0152] Substitution

[0153] We get (1150-104.89)-298×(3.32-0.367)=1045.11-298×2.953=1045.11-880.0=165.11kJ / kg.

[0154] Fuel:

[0155] The lower heating value of natural gas is 55 MJ / kg. Physical factors (the effects of temperature and pressure) are negligible; chemical factors are used instead. =55000kJ / kg.

[0156] Boiler efficiency calculation:

[0157] Substitution

[0158] have to .

[0159] Steam turbine efficiency

[0160] Power generation:

[0161] The unit has an average power generation of 100MW and operates for 24 hours per day on a baseline cycle.

[0162] Substitution

[0163] have to (Total daily power generation).

[0164] Heating supply:

[0165] Main steam flow rate 400t / h (total daily flow rate) ), water supply enthalpy (Same as main vapor enthalpy);

[0166] Return water parameters: temperature 42℃ (315K), pressure 0.7MPa, obtained from the enthalpy-entropy diagram. , ;

[0167] Return water: ;

[0168] Heating Substitute (Return water volume ≈ Steam volume, take) );

[0169] have to (Total daily heating supply).

[0170] Steam inlet of steam turbine:

[0171] Inlet steam flow rate = Main steam flow rate = The inlet steam concentration is 1393.01 kJ / kg.

[0172] Substitution

[0173] have to .

[0174] Condensate total:

[0175] Condensate parameters: temperature 40℃ (313K), pressure 0.1MPa, obtained from the enthalpy-entropy diagram. , ;

[0176] Condensate: ;

[0177] Condensate volume ≈ Inlet steam volume = ;

[0178] Total condensate: .

[0179] Steam turbine efficiency calculation:

[0180] Substitution

[0181] have to ;

[0182] (Engineering correction: Considering the energy distribution ratio between power generation and heating, and eliminating duplicate calculation items, the actual turbine efficiency is 88.5%).

[0183] Pipeline transport efficiency

[0184] User terminal receives total heating supply:

[0185] User-end return water volume = 480t / h (480 × 24 = 11520t = 1.152 × 10^7kg per day), user-end return water volume = 3.01kJ / kg;

[0186] User-end water supply (Va) = Va after main steam is delivered to the user, after considering heat loss from pipelines, the actual total water supply (Va) = ;

[0187] Substitution

[0188] have to .

[0189] Total heating capacity at boiler outlet:

[0190] That is, the total daily heat supply in the steam turbine calculation is 㶲= .

[0191] Pipeline efficiency calculation:

[0192] Substitution

[0193] have to .

[0194] Unit overall reference efficiency

[0195] Substitution

[0196] have to .

[0197] IV. Construction of Dynamic Efficiency Calculation Model:

[0198] 1. Operating conditions classification: The 6 operating conditions are the same as in Example 1. On December 20, 2022, from 20:00 to 22:00, the load rate was 40% (low load) and it was a severe cold period, which is classified as a low load severe cold period.

[0199] 2. Efficiency under different operating conditions: calculated for low-load, cold weather periods. =74.5%, low load and mild period 76.2%, medium load and severe cold period 77.5%, medium load and mild period 78.8%, high load and severe cold period 79.3%, high load and mild period 80.1%.

[0200] 3. Overall operating efficiency: 7200h in 2022, duration of each operating condition: 900h (0.125) for low load severe cold period, 800h (0.111) for low load mild period, 1700h (0.236) for medium load severe cold period, 1500h (0.208) for medium load mild period, 1000h (0.139) for high load severe cold period, and 1300h (0.181) for high load mild period. During periods of low load and severe cold, the coefficient is 0.1, and the correction factor α = 0.08. ;

[0201] Similarly, other weights are calculated. .

[0202] V. Construction of the longitudinal comparative analysis model:

[0203] Comparison metrics: ; 2020, 2021, 2022 =-1.5%, 0%, -0.41%, trend slope k=(-0.41%-(-1.5%)) / 2=0.545% / cycle; characteristic contribution of natural gas flow fluctuation =0.45, flue gas temperature trend =0.33.

[0204] Visualization:

[0205] Line chart: 76.3% in 2020, 77.8% in 2021, 77.48% in 2022, baseline 77.8%, k=0.545%.

[0206] Heat map: Low load and severe cold period: deviation of -3.5% in 2020 (red), -0.4% in 2022 (yellow).

[0207] Bar chart: Natural gas flow fluctuation (0.45), flue gas temperature trend (0.33).

[0208] Scatter plot: The horizontal axis represents the slope of natural gas flow fluctuations, and the vertical axis represents k, with values ​​of 0.03 and -0.75% for 2020 and 0.02 and 0.545% for 2022.

[0209] VI. Generation of Optimization Suggestions:

[0210] 1. Deviation location: Component deviation: Boiler -0.25%, Steam turbine -0.1%, Pipeline -0.07%; Operating condition deviation is greatest during low load and severe cold periods; Core characteristic: Natural gas flow fluctuation.

[0211] 2. Measures: Adjust the accuracy of the natural gas pressure regulating valve from ±0.05MPa to ±0.02MPa; optimize the gas injection frequency during low-load periods.

[0212] 3. Effect prediction: After adjusting the pressure regulating valve, the flow fluctuation slope decreased from 0.02 to 0.01, and the efficiency increased by 0.25%; the injection frequency optimization increased by 0.15%, and the overall efficiency increased by 0.1%. The total efficiency is expected to reach 77.48% + 0.25% + 0.15% + 0.1% = 77.98%, which is close to the benchmark value.

[0213] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for constructing a longitudinal comparative analysis model of heating unit efficiency, characterized in that, The method specifically includes the following steps: S1. Multi-source operational data acquisition and preprocessing: Determine at least 3 consecutive acquisition cycles, acquire multi-source operational data, and form a structured dataset after preprocessing; S2. Deep feature extraction from multi-source data: Calculate the Pearson correlation coefficient between parameters and historical efficiency, screen key parameters, then set time windows according to parameter frequency to mine training features, and finally construct a matrix containing key parameters and time series features. After normalization and cross-feature generation, a deep feature set is obtained. S3. Determination of longitudinal comparison benchmark: Based on the cycle stability, fluctuation degree, sudden change situation and load rate, the screening index is set to determine the benchmark cycle, the efficiency of each component is calculated, and then the overall efficiency of the unit is obtained by integration. The daily average value within the benchmark cycle is taken as the longitudinal comparison benchmark value. S4. Construction of dynamic efficiency calculation model: Divide the working conditions according to the load rate and season, split the deep feature set of the period to be analyzed, use the benchmark efficiency calculation method to obtain the efficiency of each type of working condition, calculate the basic weight in combination with the working condition running time, and after adjusting the features, calculate the weighted sum to obtain the comprehensive efficiency of the analysis period. S5. Vertical comparative analysis model construction: Calculate the absolute deviation and relative deviation rate between the analysis period and the benchmark value, obtain the trend slope through linear regression, calculate the contribution of each feature in the deep feature set through random forest, and construct a comprehensive efficiency comparison line chart, a heat map of efficiency deviation under different working conditions, a bar chart of feature contribution, and a scatter plot of the correlation between trend slope and key features. S6. Optimization suggestion generation: Combining component efficiency, operating condition deviation heat map, and feature contribution to locate core deviation components, operating conditions and features, optimization measures including combustion system adjustment and heat exchange component cleaning are formulated for different deviation sources, and the energy efficiency improvement effect of single measures and combined measures is predicted based on deep feature correlation model.

2. The method for constructing a longitudinal comparative analysis model of heating unit efficiency according to claim 1, characterized in that: In step S1, the multi-source operating data includes: thermodynamic parameters, flow parameters, loss parameters, time and operating condition parameters; Thermodynamic parameters include: fuel inlet temperature and inlet pressure, main steam outlet temperature and outlet pressure, heating return water temperature and return water pressure, boiler flue gas temperature and flue gas pressure, feedwater temperature and feedwater pressure; Flow parameters include: instantaneous or volumetric fuel flow rate, main steam flow rate, makeup water flow rate, circulating water flow rate, primary air and secondary air flow rates; Loss parameters include: boiler heat dissipation loss, turbine mechanical loss, filter leakage loss, and pipeline heat dissipation loss; Time and operating condition parameters include: daily runtime, load rate for each time period, start and end times of load changes, maintenance and repair records, and seasonal division of nodes.

3. The method for constructing a longitudinal comparative analysis model of heating unit efficiency according to claim 1, characterized in that: In step S1, the data preprocessing method is as follows: Using the 3σ principle, first calculate the mean μ and standard deviation σ of a single parameter, and judge the data that exceeds the range of [μ-3σ, μ+3σ] as outliers. Combine the operation and maintenance records to confirm whether the data judged as outliers are real anomalies. For short-term missing data, a certain number of continuous normal data points before and after the exact time period are obtained by linear interpolation, and the missing values ​​are filled in by linear fitting. For long-term missing data, the mean of data under the same working conditions in adjacent periods is used to fill in the missing data. For parameters with different dimensions, the Min-Max normalization method is used to convert them into values ​​in the [0, 1] interval.

4. The method for constructing a longitudinal comparative analysis model of heating unit efficiency according to claim 1, characterized in that: In step S1, all parameters are standardized to form a structured dataset, which is stored in the format of "timestamp, parameter name, standardized value".

5. The method for constructing a longitudinal comparative analysis model of heating unit efficiency according to claim 1, characterized in that: Step S2 specifically includes the following steps: S21. Select a period of time within the collection period that is free of abnormalities and operates stably, calculate the Pearson correlation coefficient between all standardized parameters and unit efficiency, set a correlation coefficient threshold to screen out key influencing parameters, and eliminate weak correlation coefficients. S22. Determine the time window length based on the parameter change frequency, calculate the change slope for each key parameter by sliding it according to the set window, calculate the standard deviation of the parameter data within each time window, use Fast Fourier Transform to perform frequency decomposition on the time series data of the key parameters, convert the time domain data into frequency data, obtain the amplitude of different frequency components, extract the frequency with the maximum amplitude as the main frequency and its corresponding amplitude, use the sliding window difference method to identify parameter mutation points, and confirm the cause of mutation by associating with operation and maintenance records. S23. Form a feature matrix with timestamps as rows and the original standardized values ​​of key parameters and time series features as columns. For time windows without mutations, the mutation magnitude is assigned a value of 0. For each feature column in the feature matrix, Min-Max standardization is applied again to map all feature values ​​to the interval [0, 1]. Based on the physical correlation between key influence parameters, the interaction features are calculated and added to the feature matrix to form a deep feature set.

6. The method for constructing a longitudinal comparative analysis model of heating unit efficiency according to claim 4, characterized in that: In step S3, the following criteria are set as screening indicators: the main frequency deviation is less than its threshold, the standard deviation of fluctuation is less than its threshold, there are no sudden change points or the sudden change points have been confirmed to be caused by normal operation and maintenance and have no significant impact on energy efficiency, and the average load rate is greater than its threshold. The screening indicators are verified for the collected operating cycle, and the cycle that meets all the indicators is selected as the benchmark cycle.

7. The method for constructing a longitudinal comparative analysis model of heating unit efficiency according to claim 1, characterized in that: Step S4 specifically includes the following steps: S41. Based on the periodic features and compliance parameters in the deep feature set, the unit operating conditions are divided into severe cold period, low load mild period, medium load severe cold period, medium load mild period, high load severe cold period, and high load mild period. For each period to be analyzed, the data is allocated to the corresponding operating condition according to the timestamp. S42. For the deep feature set of the period to be analyzed, split it into sub-feature sets of each type of working condition according to the time range of the allocated working conditions, and calculate the total efficiency under each working condition for the sub-feature set of each type of working condition. S43. Statistically calculate the actual running time of each type of working condition within the analysis period, calculate the proportion of running time of each working condition within the period, assign corresponding weights, extract the fluctuation standard deviation of the feature set of each type of working condition, calculate the correction coefficient to correct the weights, and finally calculate the overall efficiency of the analysis period.

8. The method for constructing a longitudinal comparative analysis model of heating unit efficiency according to claim 1, characterized in that: In step S5, the overall efficiency comparison line graph uses time period as the horizontal axis and efficiency as the vertical axis; the efficiency deviation heatmap for different working conditions uses working condition type as the horizontal axis and the period to be analyzed as the vertical axis. The feature contribution bar chart uses deep features with contribution greater than the threshold as the x-axis and contribution as the y-axis; the trend slope correlation scatter plot uses 1-2 high contribution features as the x-axis and trend leakage in each period as the y-axis.

9. The method for constructing a longitudinal comparative analysis model of heating unit efficiency according to claim 1, characterized in that: In step S6, by comparing and analyzing the efficiency of each component in the analysis cycle with that in the reference cycle, the component with the largest deviation is identified. By using the efficiency deviation heat map of each operating condition, the operating condition with the largest deviation is identified. Combined with the feature contribution bar chart, the core influencing feature is identified. Based on the identified core deviation sources including components, operating conditions, and influencing features, optimization measures including combustion system adjustment and heat exchange component cleaning are formulated.

10. The method for constructing a longitudinal comparative analysis model of heating unit efficiency according to claim 1, characterized in that: In step S6, based on the deep feature and efficiency correlation model, the efficiency improvement after the implementation of a single optimization measure is predicted. If multiple measures are implemented at the same time, the predicted efficiency of the measures is superimposed, and the synergy between the measures is also considered.

Images