Intelligent load distribution method and system for diesel generator

Abstract

The invention relates to the technical field of diesel generator control, and discloses an intelligent load distribution method and system for a diesel generator, and the method comprises the steps: collecting the temperature, vibration and noise data of a unit, extracting a temperature change rate, a vibration frequency peak value and a noise decibel mean value after preprocessing, and carrying out the standardization processing to form a characterization data set. And inputting the characteristic data into a health assessment model for state grading, and when the health grade is lower than a preset threshold value, dynamically calculating a load distribution proportion based on the characteristic parameters. The method comprises the following steps: predicting future demands by combining historical load data and adopting an LSTM network, optimizing operation parameter configuration by applying a dynamic programming algorithm, calculating an optimal parameter adjustment step length by setting a deviation threshold range of temperature, vibration and noise, and defining a deviation proportion of the set temperature, vibration and noise through a preset decision variable. And final operation parameter configuration is determined. Accurate evaluation of the health state and dynamic optimization of load distribution are achieved, and the operation efficiency of the diesel generating set is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of diesel generator control, and in particular to an intelligent load distribution method and system for a diesel generator. Background Art

[0002] Currently, diesel generators are core equipment supporting energy supply in modern industrial and civilian sectors. Their operating efficiency and stability directly impact the reliability and economic viability of power systems. With the increasing complexity of energy demand scenarios and the development of intelligent technologies, generator sets must sense multi-dimensional operating parameters such as temperature, vibration, and noise in real time under dynamic load conditions. A dynamic assessment system for unit health and load capacity, based on historical data, is then constructed to enable intelligent load distribution across multiple units. However, quantitative assessment of unit health and load capacity requires integrating the spatiotemporal correlations of multi-source heterogeneous parameters, while dynamic load demand forecasting must overcome the interference of historical data volatility and sudden changes in operating conditions. Furthermore, in scenarios with high load fluctuations or sudden power supply, balancing real-time response speed with system stability and establishing optimization strategies that balance efficiency and equipment lifespan remains a key bottleneck in achieving intelligent control. Therefore, the key issue in this study is how to accurately assess diesel generator health and load capacity by acquiring parameters such as temperature, vibration, and noise, combining them with historical data, and dynamically adjusting operating parameters to achieve intelligent load distribution.

[0003] In one existing technology, temperature, vibration, and noise sensors are deployed to collect data at fixed intervals and independently set thresholds (such as temperature > 85°C triggers load reduction). The health assessment uses a linear weighted model to superimpose the deviations of each parameter from the design value according to a fixed weight. If the comprehensive score is lower than the threshold, the rated power of the unit is reduced by 20%. The load forecast generates a demand curve based on the moving average method of historical data. The control center polls and distributes the load in order of unit number, and transfers it evenly to the standby unit in case of overload. The regulation link uses a fixed parameter PID controller to adjust the throttle opening and stabilize the voltage with a response speed of 500ms.

[0004] Existing technologies, due to their use of static methods such as fixed threshold judgments, linear weighted models, and moving average predictions, struggle to capture the nonlinear spatiotemporal correlations between multiple parameters. Isolating parameters like temperature and vibration in isolation causes health assessments to deviate from the true equipment degradation trajectory. Fixed weight allocation ignores changes in parameter sensitivity under dynamic conditions, while the moving average method cannot adapt to sudden load changes and data fluctuations. Round-robin load distribution strategies fail to couple the real-time health of the unit with its load potential. Furthermore, fixed-parameter PID control lacks compensation for system inertia hysteresis. Ultimately, these methods lead to assessment bias and response hysteresis under multi-dimensional coupled disturbances, resulting in health status misjudgments, load distribution imbalances, and dynamic regulation instability. Consequently, existing technologies are unable to accurately assess the health of diesel generators, dynamically adjust operating parameters, and achieve intelligent load distribution. Summary of the Invention

[0005] The present invention provides an intelligent load distribution method and system for diesel generators, so as to achieve accurate health status assessment and dynamic optimization of load distribution, thereby improving the operating efficiency of diesel generator sets.

[0006] In a first aspect, in order to solve the above technical problems, the present invention provides an intelligent load distribution method for a diesel generator, comprising:

[0007] Collect temperature data, vibration signals and noise intensity during the operation of the diesel generator, store them in the database, and obtain the original operating data;

[0008] Performing data preprocessing on the original operating data to obtain preprocessed data;

[0009] Extracting the temperature change rate, vibration frequency peak value and noise decibel mean from the pre-processed data, and obtaining characterization data through data standardization and characterization processing;

[0010] Inputting the characterization data into a pre-established health status assessment model to determine the health status level of the diesel generator;

[0011] When the health status level is lower than a preset health level threshold, calculating a load distribution ratio according to the temperature change rate, the vibration frequency peak value, and the noise decibel average;

[0012] Obtain historical load demands and use the long short-term memory network algorithm to determine future load demands;

[0013] According to the load distribution ratio and the future load demand, a dynamic programming algorithm is used to calculate the parameter adjustment step size, and the deviation ratios of set temperature, vibration and noise are defined through preset decision variables to determine the final operating parameter configuration.

[0014] In an optional embodiment, performing data preprocessing on the original operation data to obtain preprocessed data includes:

[0015] The temperature data is smoothed using a mean filtering algorithm to obtain a smoothed temperature;

[0016] Performing a discrete degree characteristic analysis on the vibration signal to obtain a vibration characteristic;

[0017] Identifying and eliminating abnormal vibration points in the vibration characteristics by using a standard deviation threshold method to obtain an adjusted vibration characteristic;

[0018] analyzing the vibration change trend of the adjusted vibration characteristics by fast Fourier transform to obtain vibration trend data;

[0019] Identifying and eliminating noise anomalies in the noise intensity by a standard deviation threshold method to obtain filtered noise;

[0020] The smoothed temperature, the vibration trend data, and the filtered noise are used as pre-processing data.

[0021] In an optional embodiment, the temperature change rate, vibration frequency peak value and noise decibel mean are extracted from the preprocessed data, and characterization data is obtained by performing data standardization and characterization processing, including:

[0022] Extracting the temperature change rate, the vibration frequency peak value, and the noise decibel mean from the preprocessed data, and performing data normalization processing to obtain standardized data, wherein the standardized data includes the standardized temperature change rate, the standardized vibration frequency peak value, and the standardized noise decibel mean;

[0023] According to the standardized temperature change rate, a time series distribution of the temperature change rate is calculated by a sliding window statistical method and a temperature fluctuation range is determined to obtain temperature distribution data;

[0024] Performing cluster analysis on the temperature distribution data using the K-means algorithm to obtain a temperature state classification;

[0025] Calculating a peak value variation of the normalized vibration frequency peak value, and when the peak value variation exceeds a preset peak value variation threshold, marking the corresponding normalized vibration frequency peak value as an abnormal vibration point, thereby obtaining abnormal vibration data;

[0026] Using a random forest algorithm to perform correlation analysis on the temperature state classification and the abnormal vibration data to obtain a correlation state category;

[0027] Analyzing the operating state characteristics of the diesel generator by using the Spearman rank correlation coefficient according to the associated state category and the standardized noise decibel mean;

[0028] The temperature state classification, the abnormal vibration data, the associated state category, and the operating state feature are used as characterization data.

[0029] In an optional embodiment, the training process of the health status assessment model includes:

[0030] Obtain historical characterization data and corresponding historical health levels;

[0031] The support vector machine algorithm is used to build the initial health assessment model, the classification boundary is initialized by the RBF kernel function, and the penalty coefficient is set to determine the tolerance error threshold;

[0032] Based on the historical characterization data, a kernel density estimation method is used to calculate the curvature adjustment parameter of the classification boundary, and the curvature of the classification boundary is dynamically adjusted through the RBF kernel function;

[0033] According to the historical health level, the penalty coefficient is optimized by a grid search method to determine an error tolerance threshold;

[0034] Perform cross-validation on the adjusted classification boundaries and calculate the classification accuracy and error distribution set after boundary adjustment;

[0035] When the error distribution set exceeds the error tolerance threshold, the curvature parameter of the classification boundary is re-optimized using the gradient descent method, and the classification boundary is iteratively updated until the error distribution set does not exceed the error tolerance threshold;

[0036] The optimized classification boundary and health status level are matched in multi-dimensional space through RBF kernel function mapping to determine the hyperplane equation corresponding to each health level;

[0037] Combined with the time series characteristics of the historical health levels, the sliding window method is used to extract the dynamic trend parameters of health status changes, and the dynamic trend parameters are correlated with the hyperplane equation to construct the final health status assessment model.

[0038] In an optional embodiment, when the health status level is lower than a preset health level threshold, calculating the load distribution ratio according to the temperature change rate, the vibration frequency peak value, and the noise decibel mean value includes:

[0039] Calculating the relative deviation of the temperature change rate from a preset temperature reference value, and the absolute deviation of the vibration frequency peak from a preset vibration reference value, to obtain an initial deviation data set;

[0040] Using an entropy weight method to perform weighted processing on the initial deviation data and the noise decibel mean to obtain a weighted deviation sequence;

[0041] Calculating the load distribution ratio of each generator unit using a preset load distribution ratio formula according to the weighted deviation sequence;

[0042] The load distribution ratio formula is as follows:

[0043]

[0044] Among them, P i represents the load sharing ratio of the i-th generator unit, D i and D j represents the weighted deviation value of the i-th and j-th generator units, which is obtained by weighted calculation of the initial deviation data and the noise decibel mean through the entropy weight method, ND Represents the number of data in the weighted deviation sequence, and ∈ represents the preset proportional error coefficient.

[0045] In an optional embodiment, obtaining historical load demands and determining future load demands using a long short-term memory network algorithm includes:

[0046] Obtaining historical load demands, and extracting periodic characteristics of the historical load demands using a time series decomposition algorithm;

[0047] Constructing a long short-term memory network model, setting a preset number of hidden layer units and time step, inputting the periodic features into the long short-term memory network model for training, and outputting a load demand change trend;

[0048] Based on the load demand change trend and the historical load demand fluctuation range statistics, the quantile regression algorithm is used to calculate the forecast range of future load demand;

[0049] The Holt-Winters triple exponential smoothing method is used to extract long-term change characteristics of the forecast interval;

[0050] The long-term variation characteristics are matched with the spatiotemporal characteristics of the periodic characteristics by a dynamic time warping algorithm, and a weighted fusion method is used to generate future load demands.

[0051] In an optional embodiment, the parameter adjustment step size is calculated using a dynamic programming algorithm based on the load distribution ratio and the future load demand, the deviation ratio of the set temperature, vibration and noise is defined by preset decision variables, and the final operating parameter configuration is determined, including:

[0052] Based on the future load demand, calculating a parameter adjustment step size of the generator operating parameters by a dynamic programming algorithm;

[0053] Calculate the deviation ratio data of the temperature change rate, the vibration frequency peak value, and the noise decibel mean value according to the parameter adjustment step size and the preset decision variable definition rule;

[0054] Calculating an operating parameter configuration according to the deviation ratio data and the load distribution ratio;

[0055] The operating parameter configuration is calculated using the following formula:

[0056]

[0057] Among them, C i represents the final operating parameter configuration value of the i-th generator unit, P i represents the load distribution ratio of the i-th generator unit, Indicates the total load regulation coefficient of the system, ω k and k∈{T,V,N}, that is, ω T ,ω V and ω N Denotes the deviation weight coefficients of temperature, vibration and noise, which are obtained by the decision variable definition rules. Right now and It represents the deviation ratio of the temperature change rate, vibration frequency peak value, and noise decibel mean of the i-th unit, C base_i Represents the reference value of the rated operating parameters of the i-th generator.

[0058] In a second aspect, the present invention provides an intelligent load distribution system for a diesel generator, comprising:

[0059] The data acquisition module is used to collect temperature data, vibration signals and noise intensity during the operation of the diesel generator, store them in the database, and obtain the original operating data;

[0060] A data preprocessing module, configured to perform data preprocessing on the original operation data to obtain preprocessed data;

[0061] A data characterization module is used to extract the temperature change rate, vibration frequency peak value and noise decibel mean from the pre-processed data, and obtain characterization data through data standardization and characterization processing;

[0062] A health assessment module, configured to input the characterization data into a pre-established health status assessment model to determine the health status level of the diesel generator;

[0063] a load ratio calculation module, configured to calculate a load distribution ratio according to the temperature change rate, the vibration frequency peak value, and the noise decibel mean when the health status level is lower than a preset health level threshold;

[0064] The future load forecasting module is used to obtain historical load demands and determine future load demands using a long short-term memory network algorithm;

[0065] An operating parameter calculation module is used to calculate the parameter adjustment step size using a dynamic programming algorithm based on the load distribution ratio and the future load demand, define the deviation ratio of the set temperature, vibration and noise through preset decision variables, and determine the final operating parameter configuration.

[0066] In a third aspect, the present invention also provides an electronic device comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the intelligent load distribution method for the diesel generator described in any one of the above is implemented.

[0067] In a fourth aspect, the present invention also provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the above-mentioned intelligent load distribution methods for diesel generators.

[0068] Compared with the prior art, the present invention has the following beneficial effects:

[0069] (1) By integrating the multi-dimensional feature data of temperature, vibration, and noise and combining dynamic clustering and association analysis technology, the limitations of traditional single-parameter threshold detection are overcome, and a refined assessment of the health status of diesel generators is achieved, significantly improving the accuracy and robustness of health status determination.

[0070] (2) Based on the support vector machine, a dynamically adjustable classification boundary model is constructed. Through kernel function curvature optimization and time series feature extraction, the nonlinear change characteristics in the equipment degradation process are effectively captured, and the adaptability and stability of the health assessment model under complex working conditions are significantly enhanced.

[0071] (3) The entropy weight method is used to fuse multi-dimensional deviation parameters, and the dynamic programming algorithm is combined to build a load distribution optimization model. Through the dynamic compensation mechanism of deviation weights, the rapid response and balanced distribution of multiple units can be achieved in the load mutation scenario, significantly improving the dynamic adjustment capability of the system.

[0072] (4) Combining the LSTM network with spatiotemporal feature matching technology, through multi-scale feature decomposition and fusion prediction mechanism, it can effectively capture the periodic laws and sudden fluctuation characteristics of load demand, and significantly improve the reliability and anti-interference ability of future load forecasting.

[0073] In summary, this invention builds a complete intelligent control system from state perception to dynamic optimization: It accurately quantifies equipment health status through multi-dimensional data fusion and dynamic classification models; optimizes multi-unit coordinated load distribution based on an entropy-weighted dynamic compensation mechanism; improves load prediction accuracy by combining deep learning with spatiotemporal feature analysis; and ultimately achieves coordinated optimization of system efficiency and equipment lifespan through global parameter optimization. This solution addresses the technical shortcomings of traditional methods, such as one-sided health assessment, rigid load distribution, and insufficient prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 This is a flow chart of a method for intelligent load distribution of a diesel generator provided by a first embodiment of the present invention;

[0075] Figure 2 2 is a schematic structural diagram of an intelligent load distribution system for a diesel generator provided in accordance with a second embodiment of the present invention. DETAILED DESCRIPTION

[0076] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0077] Reference Figure 1 The first embodiment of the present invention provides a method for intelligent load distribution of a diesel generator, comprising the following steps:

[0078] S11, collecting temperature data, vibration signals and noise intensity of the diesel generator during operation, storing them in a database, and obtaining original operation data;

[0079] S12, performing data preprocessing on the original operation data to obtain preprocessed data;

[0080] S13, extracting the temperature change rate, vibration frequency peak value and noise decibel mean from the preprocessed data, and obtaining characterization data through data standardization and characterization processing;

[0081] S14, inputting the characterization data into a pre-established health status assessment model to determine the health status level of the diesel generator;

[0082] S15, when the health status level is lower than a preset health level threshold, calculating a load distribution ratio according to the temperature change rate, the vibration frequency peak value, and the noise decibel average;

[0083] S16, obtains historical load demands and uses a long short-term memory network algorithm to determine future load demands;

[0084] S17, according to the load distribution ratio and the future load demand, a dynamic programming algorithm is used to calculate the parameter adjustment step size, the deviation ratio of the set temperature, vibration and noise is defined through preset decision variables, and the final operating parameter configuration is determined.

[0085] In step S11, the temperature data, vibration signal and noise intensity of the diesel generator during operation are collected and stored in a database to obtain original operation data.

[0086] Specifically, data collection is achieved through a sensor network installed at key locations on the diesel generator. Temperature sensors are placed in core heat-generating areas of the equipment, such as the cylinder surface and cooling line nodes, to continuously record thermal changes between metal components and fluids. Vibration sensors are fixed at the joints of the mechanical transmission structure to collect the dynamic mechanical response of rotating components and supporting structures. Noise sensors are distributed in the equipment's periphery to capture the propagation characteristics of sound waves in the air. The sensors convert analog signals into digital signals via an industrial bus protocol and transmit them to a central controller. A synchronization mechanism ensures the time alignment of multi-source data. For example, one type of vibration sensor uses a triaxial acceleration measurement mode, capable of simultaneously capturing mechanical vibration waveforms in the lateral, longitudinal, and vertical directions. The noise monitoring unit periodically records sound pressure level data during equipment operation, generating an acoustic signature curve. The collected raw data is stored in a structured time series database containing fields for device identification, timestamps, and physical quantity values, forming a basic dataset reflecting the comprehensive operating status of the equipment. This step establishes a three-dimensional monitoring system covering thermodynamics, mechanical dynamics, and acoustic characteristics, providing multi-dimensional, highly synchronized raw information input for subsequent condition assessment and load optimization.

[0087] In step S12, data preprocessing is performed on the original operating data to obtain preprocessed data.

[0088] In a specific embodiment, preprocessing the original operating data to obtain preprocessed data includes:

[0089] The temperature data is smoothed using a mean filtering algorithm to obtain a smoothed temperature;

[0090] Performing a discrete degree characteristic analysis on the vibration signal to obtain a vibration characteristic;

[0091] Identifying and eliminating abnormal vibration points in the vibration characteristics by using a standard deviation threshold method to obtain an adjusted vibration characteristic;

[0092] analyzing the vibration change trend of the adjusted vibration characteristics by fast Fourier transform to obtain vibration trend data;

[0093] Identifying and eliminating noise anomalies in the noise intensity by a standard deviation threshold method to obtain filtered noise;

[0094] The smoothed temperature, the vibration trend data, and the filtered noise are used as pre-processing data.

[0095] Specifically, temperature data is smoothed using a mean filter algorithm, which selects data points over a fixed time span and calculates the arithmetic mean, effectively filtering out random fluctuations caused by sensor noise. For example, when a 5-second time window is set, the system averages the temperature values at the current moment and the two seconds before and after, eliminating the impact of transient interference on trend analysis.

[0096] When performing feature analysis on a vibration signal, the first step is to calculate the dispersion characteristics of the vibration data. This involves evaluating the deviation of each data point from the average value of the entire data set. Specifically, the average value of all vibration data points is determined, and then the difference between each data point and this average value is calculated. Based on these differences, the dispersion of the entire vibration signal is quantified, thereby obtaining the vibration signature.

[0097] Next, we identify and remove outliers from the vibration signature by setting a threshold based on a multiple of the standard deviation. This threshold is based on the previously calculated dispersion of the vibration data and sets a limit for determining anomalies. Any data point that deviates from the mean by more than this limit is considered an outlier and removed from the dataset. This effectively removes extreme values that may be due to measurement error or unexpected system conditions, ensuring that subsequent analysis is based on a more accurate and reliable dataset—that is, a modified vibration signature.

[0098] Subsequently, a Fast Fourier Transform (FFT) was used to analyze the changing trends of the adjusted vibration signature. The Fast Fourier Transform is a method that converts time-domain signals into frequency-domain signals, allowing the characteristics of the vibration signal to be observed from a frequency perspective. By applying the Fast Fourier Transform to the adjusted vibration signature, the original time series data can be decomposed to reveal its underlying frequency components. This means that it is possible to identify which frequency components dominate the vibration signal and how their intensity changes over time.

[0099] Noise data processing utilizes a dual filtering mechanism. First, the statistical distribution parameters of the noise decibel value are calculated, and a dynamic threshold based on the mean and standard deviation is established. When noise exceeding a preset decibel level is detected, the system automatically identifies it as electromagnetic interference or transient impact noise and replaces it with the median value of the adjacent valid data. This preprocessed noise sequence retains the inherent acoustic characteristics of the equipment and eliminates interference from sudden ambient noise.

[0100] The entire preprocessing process forms a standardized data stream. For example, after discrete analysis of the raw vibration data of a certain type of generator set, abnormal high-frequency vibration components caused by bearing wear were successfully identified. Through signal enhancement and anomaly suppression, this stage transforms the raw data into a characteristic sequence that reflects the equipment's true operating status, providing high-quality input for subsequent health assessments.

[0101] In step S13, the temperature change rate, vibration frequency peak value and noise decibel mean are extracted from the pre-processed data, and characterization data is obtained through data standardization and characterization processing.

[0102] In a specific embodiment, the temperature change rate, vibration frequency peak value and noise decibel mean are extracted from the preprocessed data, and the characterization data are obtained by data standardization and characterization processing, including:

[0103] Extracting the temperature change rate, the vibration frequency peak value, and the noise decibel mean from the preprocessed data, and performing data normalization processing to obtain standardized data, wherein the standardized data includes the standardized temperature change rate, the standardized vibration frequency peak value, and the standardized noise decibel mean;

[0104] According to the standardized temperature change rate, a time series distribution of the temperature change rate is calculated by a sliding window statistical method and a temperature fluctuation range is determined to obtain temperature distribution data;

[0105] Performing cluster analysis on the temperature distribution data using the K-means algorithm to obtain a temperature state classification;

[0106] Calculating a peak value variation of the normalized vibration frequency peak value, and when the peak value variation exceeds a preset peak value variation threshold, marking the corresponding normalized vibration frequency peak value as an abnormal vibration point, thereby obtaining abnormal vibration data;

[0107] Using a random forest algorithm to perform correlation analysis on the temperature state classification and the abnormal vibration data to obtain a correlation state category;

[0108] Analyzing the operating state characteristics of the diesel generator by using the Spearman rank correlation coefficient according to the associated state category and the standardized noise decibel mean;

[0109] The temperature state classification, the abnormal vibration data, the associated state category, and the operating state feature are used as characterization data.

[0110] Specifically, the system first extracts features and standardizes the pre-processed temperature, vibration, and noise data to eliminate the impact of dimensional differences on the model. The temperature change rate is obtained by calculating the slope of adjacent time points. For example, if the temperature rises from 80°C to 82°C in a certain period of time in 10 seconds, the change rate is 0.2°C / s. The vibration frequency peak extracts the frequency value corresponding to the maximum energy from the spectrum analysis results, such as the 125Hz component with the largest amplitude. The noise decibel mean takes the average sound pressure level within the sliding window, such as the noise is maintained in the range of 72-75dB within a 5-minute window.

[0111] Data normalization uses mean-variance normalization to transform raw feature values into distributions with zero mean and unit variance. For example, if the raw data for a generator's temperature change rate has a mean of 0.15°C / s and a standard deviation of 0.03, and a single measured value is 0.21°C / s, the normalized value is (0.21 - 0.15) / 0.03 = 2. This process makes features of different dimensions comparable, for example, a normalized vibration frequency peak of 2.5 is on the same order of magnitude as a noise decibel mean of -1.3.

[0112] Temperature fluctuation analysis utilizes a sliding window statistical method, with the system automatically setting the window length to match the equipment's operating cycle. For a typical diesel generator set, a 30-minute window is selected to analyze the distribution of the temperature change rate. If 95% of the data points are concentrated within the standardized unit interval of [-0.5, 0.7], the system determines that the temperature is stable. When the standard deviation of the data within the window exceeds 1.5 times the historical baseline, an adaptive window length adjustment mechanism is triggered to ensure effective extraction of fluctuation characteristics.

[0113] The K-means clustering algorithm categorizes temperature states into three typical categories: steady-state operation, mild fluctuations, and drastic changes. Cluster centers are determined through historical data analysis. For example, for a particular model, cluster centers correspond to standardized values of -0.3, 0.2, and 1.8. If the Euclidean distance to the nearest neighbor cluster center of a new data point falls into the drastic change category, the system flags it as an abnormal thermal state.

[0114] Vibration anomaly detection utilizes a dual-judgment mechanism, considering both the absolute value of the peak value and the rate of change. Threshold settings are based on the characteristics of the equipment model. For example, for a certain type of unit, short-term fluctuations in the peak vibration frequency should not exceed 30% of the historical average. If a peak value suddenly increases from 120Hz to 160Hz over a period of time, with a rate of change exceeding 5Hz / minute, this is considered a sign of abnormal bearing wear. The system establishes a database of abnormal vibration events and continuously optimizes threshold settings.

[0115] The random forest algorithm constructs a multidimensional feature association model, taking as input temperature status categories and vibration anomaly markers and outputting an association strength indicator. For example, analysis found that when the temperature fluctuates dramatically, the probability of vibration anomalies increases to 78%, significantly higher than the 12% during steady-state operation. The model determines association rules by calculating information gain between features. For example, if the splitting conditions for a node are temperature category > 1 and the vibration peak change rate > 4 Hz / min, the associated status is classified as high risk.

[0116] Spearman correlation analysis reveals the dynamic relationship between noise characteristics and complex states. The system calculates the rank-order correlation coefficient between the mean noise decibel value and the associated state level. For example, under certain operating conditions, the correlation coefficient reached 0.82, indicating a strong positive correlation between noise level and the deterioration of the equipment's overall state. The feature fusion stage encodes temperature categories as one-hot vectors, converts vibration anomaly markers into binary features, and treats the associated state levels as ordered variables, ultimately forming a structured dataset containing 12-dimensional feature vectors. This feature space comprehensively characterizes the health status of the equipment's operation, providing a quantitative basis for subsequent intelligent decision-making.

[0117] In step S14, the characterization data is input into a pre-established health status assessment model to determine the health status level of the diesel generator.

[0118] In a specific embodiment, the training process of the health status assessment model includes:

[0119] Obtain historical characterization data and corresponding historical health levels;

[0120] The support vector machine algorithm is used to build the initial health assessment model, the classification boundary is initialized by the RBF kernel function, and the penalty coefficient is set to determine the tolerance error threshold;

[0121] Based on the historical characterization data, a kernel density estimation method is used to calculate the curvature adjustment parameter of the classification boundary, and the curvature of the classification boundary is dynamically adjusted through the RBF kernel function;

[0122] According to the historical health level, the penalty coefficient is optimized by a grid search method to determine an error tolerance threshold;

[0123] Perform cross-validation on the adjusted classification boundaries and calculate the classification accuracy and error distribution set after boundary adjustment;

[0124] When the error distribution set exceeds the error tolerance threshold, the curvature parameter of the classification boundary is re-optimized using the gradient descent method, and the classification boundary is iteratively updated until the error distribution set does not exceed the error tolerance threshold;

[0125] The optimized classification boundary and health status level are matched in multi-dimensional space through RBF kernel function mapping to determine the hyperplane equation corresponding to each health level;

[0126] Combined with the time series characteristics of the historical health levels, the sliding window method is used to extract the dynamic trend parameters of health status changes, and the dynamic trend parameters are correlated with the hyperplane equation to construct the final health status assessment model.

[0127] Specifically, the health status assessment model achieves precise health classification through multi-dimensional feature fusion and dynamic boundary optimization. Model training first integrates characteristic data from historical operating cycles, including temperature fluctuation patterns, vibration anomalies, and noise-related indicators, and labels the corresponding equipment health levels. A support vector machine algorithm constructs an initial classification model using a radial basis function kernel. The kernel width parameter is automatically initialized based on the feature space distribution density, forming a preliminary classification hyperplane.

[0128] Kernel density estimation technology analyzes the clustering characteristics of historical data in feature space and dynamically adjusts the curvature parameter of the kernel function. When a nonlinear distribution of health status data is detected, the system automatically increases the kernel function width in the corresponding region to better align the classification boundaries with the data distribution. For example, to address the coupling of temperature fluctuations and vibration anomalies common in aging units, adjusting the kernel function curvature can create a curved interface that fits the data distribution.

[0129] The grid search method systematically traverses combinations of penalty coefficients and error tolerance thresholds within a preset parameter space, selecting the optimal parameter configuration through 50-fold cross-validation. When the historical data for a particular unit shows a high rate of misclassification of Level III health status, the optimized penalty coefficient increases the weight of such samples, forcing the classification boundary to shift toward high-density areas in the feature space, thereby reducing the risk of misclassification of critical levels.

[0130] During the iterative optimization phase of the model, a gradient descent algorithm is used to dynamically correct the classification boundaries. When the validation set error exceeds the tolerance threshold, the system calculates the gradient of the classification loss with respect to the kernel function parameters and updates the parameters along the negative gradient until the error converges. For example, if a vibration anomaly feature is too sensitive to the classification boundary during a certain iteration, the gradient update can be used to reduce the weight of this feature in the kernel function, allowing the model to more evenly consider the impact of multi-dimensional features.

[0131] The hyperplane matching stage maps the optimized classification model to the health level space, establishing a corresponding multidimensional geometric interface for each health level. The system determines the degree of membership by calculating the relative distance of new data points to each hyperplane. For example, if a sample has the shortest Mahalanobis distance to the Level II health hyperplane and meets the threshold conditions, the current status is determined to be Level II. The time series analysis module extracts the evolution trend of the device health status. For example, if the health level migrates from Level I to Level II over three consecutive inspection cycles, the dynamic trend parameters will modify the current classification results, providing an early warning of potential health degradation risks.

[0132] In practical applications, the trained model receives real-time feature data streams, automatically calculates the matching index for each health level, and conducts comprehensive analysis based on recent state transition trends. For example, if a unit's current features match the Level II health hyperplane, but the associated state level has continuously declined over the past two hours, the system will trigger a preventive maintenance alert, enabling dynamic and accurate assessment of the equipment's health status. By integrating static feature classification with dynamic trend analysis, this model improves the reliability of state assessments under complex operating conditions.

[0133] In step S15 , when the health status level is lower than a preset health level threshold, a load distribution ratio is calculated according to the temperature change rate, the vibration frequency peak value, and the noise decibel average.

[0134] In a specific embodiment, when the health status level is lower than a preset health level threshold, calculating the load distribution ratio according to the temperature change rate, the vibration frequency peak value, and the noise decibel average includes:

[0135] Calculating the relative deviation of the temperature change rate from a preset temperature reference value, and the absolute deviation of the vibration frequency peak from a preset vibration reference value, to obtain an initial deviation data set;

[0136] Using an entropy weight method to perform weighted processing on the initial deviation data and the noise decibel mean to obtain a weighted deviation sequence;

[0137] Calculating the load distribution ratio of each generator unit using a preset load distribution ratio formula according to the weighted deviation sequence;

[0138] The load distribution ratio formula is as follows:

[0139]

[0140] Among them, P i represents the load sharing ratio of the i-th generator unit, D i and D j represents the weighted deviation value of the i-th and j-th generator units, which is obtained by weighted calculation of the initial deviation data and the noise decibel mean through the entropy weight method, N D Represents the number of data in the weighted deviation sequence, and ∈ represents the preset proportional error coefficient.

[0141] Specifically, to assess the difference between the system's current state and its ideal state, we first calculate the relative deviation of the temperature change rate from a preset temperature baseline, as well as the absolute deviation of the vibration frequency peak from a preset vibration baseline. This step provides the foundational data for subsequent analysis and is referred to as the initial deviation dataset. This considers the difference between the actual measured values and the ideally expected values, quantifying the degree of deviation.

[0142] Next, the entropy weighting method is used to weight these initial deviation data and the mean noise decibel value. The entropy weighting method determines the weights of individual indicators based on information entropy theory. By calculating the information entropy of each indicator, its importance in the overall evaluation system is determined. This process normalizes the initial deviation data set and the mean noise decibel value, then calculates the information entropy of each data item, which serves as its weight. Finally, these weights are used to weight the initial deviation data and the mean noise decibel value, resulting in a weighted deviation sequence. This sequence reflects the relative condition of each generator unit after comprehensively considering temperature, vibration, and noise factors.

[0143] Finally, based on the weighted deviation sequence, the load distribution ratio of each generator unit is calculated using a preset load distribution ratio formula. This calculation process aims to balance the workload of each generator unit and avoid failures caused by overload of a unit. Specifically, for each generator unit, the inverse of its weighted deviation value is calculated, and a very small positive number (called the proportional error coefficient) is added to prevent division by zero errors. Then, this result of the generator unit is divided by the sum of the corresponding results of all generator units to obtain the load distribution ratio of the unit. This method ensures that those generator units whose operating status is closer to the ideal (that is, the weighted deviation value is lower) will bear more load, while those units with poorer performance are assigned fewer tasks, so as to achieve the purpose of optimizing the performance of the entire system. Such a strategy helps to improve the reliability and efficiency of the system, and also helps to extend the service life of the equipment.

[0144] In step S16, historical load demands are obtained, and the long short-term memory network algorithm is used to determine future load demands.

[0145] In a specific embodiment, obtaining historical load demands and determining future load demands using a long short-term memory network algorithm includes:

[0146] Obtaining historical load demands, and extracting periodic characteristics of the historical load demands using a time series decomposition algorithm;

[0147] Constructing a long short-term memory network model, setting a preset number of hidden layer units and time step, inputting the periodic features into the long short-term memory network model for training, and outputting a load demand change trend;

[0148] Based on the load demand change trend and the historical load demand fluctuation range statistics, the quantile regression algorithm is used to calculate the forecast range of future load demand;

[0149] The Holt-Winters triple exponential smoothing method is used to extract long-term change characteristics of the forecast interval;

[0150] The long-term variation characteristics are matched with the spatiotemporal characteristics of the periodic characteristics by a dynamic time warping algorithm, and a weighted fusion method is used to generate future load demands.

[0151] Specifically, in step S16, in order to obtain a forecast of future load demand, the entire process starts with processing historical load demand data, and gradually extracts features, trains models, and finally generates a forecast through a series of algorithms.

[0152] First, historical load demand is obtained and a time series decomposition algorithm is used to extract cyclical features from it. This algorithm breaks the original time series into three main components: a trend component, a seasonal component, and a residual component. The trend component is identified by smoothing the time series, which helps to understand the overall trend of the data over time. After removing the trend component, the remaining component is analyzed to detect recurring patterns or cyclical fluctuations, thereby extracting the seasonal component. The residual component represents fluctuations that are neither trend nor seasonal and is generally considered noise. This process ensures that both cyclical and trend changes in load demand are accurately captured.

[0153] Next, a long short-term memory (LSTM) model is constructed, with a preset number of hidden layer units and time step size. The LSTM model is particularly well-suited for processing time series data because it can learn and remember long-term dependencies. The model is trained using the previously extracted periodic features as input data. The model learns these features to predict future load demand trends. This neural network-based approach enables the model to identify complex patterns in load demand over time and output forecasts for future load demand.

[0154] Based on the load demand trend output by the LSTM model and the fluctuation range statistics of historical load demand data, a quantile regression algorithm is used to calculate the forecast range for future load demand. This quantile regression algorithm not only focuses on the average trend but also considers the expected load demand at different probability levels, ensuring that the forecast range covers the possible range of variation. This approach provides a more comprehensive basis for risk assessment and helps understand how load demand may change in the future.

[0155] The quantile regression algorithm first considers different quantiles, which represent the expected load demand at different probability levels. Unlike traditional regression methods, which primarily focus on the conditional mean of the dependent variable, quantile regression estimates the conditional distribution of the dependent variable at different quantiles. This means that regression analysis can be performed separately for low, medium, and high quantiles, resulting in prediction intervals for load demand at different confidence levels.

[0156] In the implementation, multiple quantiles are first determined based on historical load demand data, such as the 5%, 50%, and 95% quantiles, corresponding to low, intermediate, and high load demand levels, respectively. Then, for each selected quantile, a corresponding quantile regression model is constructed and trained. This model estimates parameters by minimizing the sum of weighted absolute residuals, with weights determined based on the selected quantile. This yields a set of parameter estimates for each quantile, describing the load demand behavior at that quantile.

[0157] Next, the trained quantile regression model is used to forecast future load demand. Because models are built for multiple quantiles, a series of forecast values are generated, which together form a prediction interval for future load demand. For example, at a 90% confidence level, the upper and lower bounds of the prediction interval can be defined by the forecast values at the 5% and 95% quantiles. This approach not only provides an estimate of the average expected load demand but also reveals the potential range of fluctuations, enabling decision makers to more fully understand the uncertainty of future load demand.

[0158] In this way, the quantile regression algorithm effectively complements traditional forecasting methods, especially when dealing with data exhibiting significant volatility and uncertainty. It allows for more nuanced and accurate predictions of future load demand, supporting more effective resource planning and risk control strategies. The entire process emphasizes the importance of understanding and forecasting time series data from multiple perspectives to improve decision-making quality.

[0159] The Holt-Winters triple exponential smoothing method is then used to process the resulting forecast intervals to extract long-term variation characteristics. This method reflects time series changes by updating the three components: level, trend, and seasonality. Initial estimates and smoothing parameters are first determined, and then the estimates of these three components are continuously adjusted based on new observations to reveal long-term trends in load demand. This method is well-suited for datasets containing significant trend and seasonal components, effectively extracting long-term variation characteristics from historical data.

[0160] Finally, the dynamic time warping (DTW) algorithm is used to match the spatiotemporal correlations between the long-term variation features extracted by the Holt-Winters method and the previously observed periodic features. The DTW algorithm ensures a more accurate reflection of the true variation pattern of load demand by finding the optimal correspondence between the two features. On this basis, a weighted fusion method is used to combine these two features to generate a final forecast of future load demand. This approach comprehensively considers periodicity, trends, and potential fluctuations, not only improving forecast accuracy but also providing a solid foundation for responding to future changes. The entire process demonstrates the importance of combining multiple algorithms and techniques to achieve accurate load demand forecasts, which helps optimize resource allocation and improve system operational efficiency.

[0161] In step S17, according to the load distribution ratio and the future load demand, a dynamic programming algorithm is used to calculate the parameter adjustment step size, the deviation ratio of the set temperature, vibration and noise is defined by the preset decision variables, and the final operating parameter configuration is determined.

[0162] In a specific embodiment, the parameter adjustment step size is calculated using a dynamic programming algorithm based on the load distribution ratio and the future load demand, the deviation ratio of the set temperature, vibration and noise is defined by preset decision variables, and the final operating parameter configuration is determined, including:

[0163] Based on the future load demand, calculating a parameter adjustment step size of the generator operating parameters by a dynamic programming algorithm;

[0164] Calculate the deviation ratio data of the temperature change rate, the vibration frequency peak value, and the noise decibel mean value according to the parameter adjustment step size and the preset decision variable definition rule;

[0165] Calculating an operating parameter configuration according to the deviation ratio data and the load distribution ratio;

[0166] The operating parameter configuration is calculated using the following formula:

[0167]

[0168] Among them, C i represents the final operating parameter configuration value of the i-th generator unit, P i represents the load distribution ratio of the i-th generator unit, Indicates the total load regulation coefficient of the system, ω k and k∈{T,V,N}, that is, ω T ,ω V and ω N Denotes the deviation weight coefficients of temperature, vibration and noise, which are obtained by the decision variable definition rules. Right now and It represents the deviation ratio of the temperature change rate, vibration frequency peak value, and noise decibel mean of the i-th unit, C base_i Represents the reference value of the rated operating parameters of the i-th generator.

[0169] Specifically, during the operating parameter configuration process, the system achieves precise control through a collaborative approach of multi-stage optimization and dynamic rules. First, based on the time periods of future load demand, a dynamic programming algorithm constructs a multidimensional state space containing temperature, vibration, and noise parameters. Combined with the physical model of the equipment, state transition equations are established to simulate the dynamic impact of different load adjustment actions on operating parameters. The algorithm reversely calculates the optimal adjustment step size for each time period, minimizing the combined costs of parameter fluctuations, supply-demand deviations, and equipment losses while meeting load demand, thereby generating a globally optimized parameter adjustment sequence.

[0170] Parameter adjustment steps are converted into standardized deviation ratio data using decision variable definition rules. The rule base pre-sets equipment model characteristic parameters and historical operating thresholds, mapping the algorithm's output adjustment step to the [0, 1] range. When the adjustment step is within a safe range, it is converted to a deviation value using a linear ratio; when it approaches a critical threshold, a nonlinear compression function is activated to prevent over-limit risks. For example, the vibration adjustment step of a certain type of generator is mapped to a deviation ratio at a 1:1 ratio within the 0-2Hz range. A logarithmic function is used to suppress the increase above 2Hz. The rule base also integrates a real-time status perception module. When equipment aging characteristics are detected, the historical threshold is automatically relaxed by 10%-15% to accommodate performance degradation.

[0171] The deviation ratio calculation module incorporates a dynamic weighting mechanism. Using an entropy weighting method, it analyzes the real-time data fluctuations of temperature, vibration, and noise parameters, dynamically assigning compensation weights to each parameter. Under high-temperature conditions, the information entropy of the temperature parameter decreases, and the system automatically increases its weight to 50%, strengthening temperature control priority. The multi-parameter coupling analysis unit calculates the correlation coefficient between parameters in real time. When a strong positive correlation is observed between vibration and noise, a joint control mode is activated, combining the weighted deviation ratios of the two to avoid the risk of resonance caused by independent adjustments.

[0172] It should be noted that is a dimensionless parameter, Limited range

[0173] Operating parameter configuration strictly follows a multi-dimensional integrated calculation formula: the load distribution ratio reflects the unit's health level, the total load adjustment coefficient ensures system-level supply and demand balance, the deviation ratio data indicates the need for parameter adjustment, and the dynamic weight coefficient determines the control priority of each parameter. The formula uses normalization to adapt the load distribution ratio to the current system's total load demand. The basic parameter values are then dynamically corrected through multi-dimensional deviation compensation terms. The final output is a precise configuration that meets both equipment health constraints and power supply requirements, achieving continuous adaptive adjustment of load distribution and equipment protection.

[0174] Reference Figure 2 A second embodiment of the present invention provides an intelligent load distribution system for a diesel generator, comprising:

[0175] The data acquisition module is used to collect temperature data, vibration signals and noise intensity during the operation of the diesel generator, store them in the database, and obtain the original operating data;

[0176] A data preprocessing module, configured to perform data preprocessing on the original operation data to obtain preprocessed data;

[0177] A data characterization module is used to extract the temperature change rate, vibration frequency peak value and noise decibel mean from the pre-processed data, and obtain characterization data through data standardization and characterization processing;

[0178] A health assessment module, configured to input the characterization data into a pre-established health status assessment model to determine the health status level of the diesel generator;

[0179] a load ratio calculation module, configured to calculate a load distribution ratio according to the temperature change rate, the vibration frequency peak value, and the noise decibel mean when the health status level is lower than a preset health level threshold;

[0180] The future load forecasting module is used to obtain historical load demands and determine future load demands using a long short-term memory network algorithm;

[0181] An operating parameter calculation module is used to calculate the parameter adjustment step size using a dynamic programming algorithm based on the load distribution ratio and the future load demand, define the deviation ratio of the set temperature, vibration and noise through preset decision variables, and determine the final operating parameter configuration.

[0182] It should be noted that the intelligent load distribution device for a diesel generator provided in an embodiment of the present invention is used to execute all the process steps of the intelligent load distribution method for a diesel generator in the above embodiment. The working principles and beneficial effects of the two correspond one to one, and therefore will not be repeated.

[0183] An embodiment of the present invention further provides an electronic device. The electronic device includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a program for intelligent load distribution of a diesel generator. When the processor executes the computer program, the steps of the above-mentioned embodiments of the intelligent load distribution method for diesel generators are implemented, such as Figure 1 Alternatively, when the processor executes the computer program, the functions of the modules / units in the above-mentioned device embodiments are realized, such as the intelligent load distribution module of the diesel generator.

[0184] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, and the instruction segments are used to describe the execution process of the computer program in the electronic device.

[0185] The electronic device may be a computing device such as a desktop computer, notebook, PDA, or smart tablet. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will appreciate that the aforementioned components are merely examples of electronic devices and do not constitute a limitation of the electronic device. The electronic device may include more or fewer components than those described above, or a combination of certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, and the like.

[0186] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the electronic device, connecting various parts of the entire electronic device using various interfaces and lines.

[0187] The memory can be used to store the computer programs and / or modules, and the processor realizes various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function (such as a sound playback function, an image playback function, etc.); the data storage area can store data created based on the use of the mobile phone (such as audio data, a phone book, etc.). In addition, the memory can include a high-speed random access memory and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0188] Wherein, if the module / unit integrated in the electronic device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the process in the above-mentioned embodiment method, and can also be completed by a computer program to instruct the relevant hardware. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, it can implement the steps of each of the above-mentioned method embodiments. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0189] It should be noted that the device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed across multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. In addition, in the drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art can understand and implement the present invention without inventive effort.

[0190] The specific embodiments described above further illustrate the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for intelligent load distribution of a diesel generator, characterized in that: include: Collect temperature data, vibration signals and noise intensity during the operation of the diesel generator, store them in the database, and obtain the original operating data; Performing data preprocessing on the original operating data to obtain preprocessed data; Extracting the temperature change rate, vibration frequency peak value and noise decibel mean from the pre-processed data, and obtaining characterization data through data standardization and characterization processing; Inputting the characterization data into a pre-established health status assessment model to determine the health status level of the diesel generator; When the health status level is lower than a preset health level threshold, calculating a load distribution ratio according to the temperature change rate, the vibration frequency peak value, and the noise decibel average; Obtain historical load demands and use the long short-term memory network algorithm to determine future load demands; According to the load distribution ratio and the future load demand, a dynamic programming algorithm is used to calculate the parameter adjustment step size, and the deviation ratios of set temperature, vibration and noise are defined through preset decision variables to determine the final operating parameter configuration.

2. The intelligent load distribution method for diesel generators according to claim 1, characterized in that: The performing data preprocessing on the original operation data to obtain preprocessed data includes: The temperature data is smoothed using a mean filtering algorithm to obtain a smoothed temperature; Performing a discrete degree characteristic analysis on the vibration signal to obtain a vibration characteristic; Identifying and eliminating abnormal vibration points in the vibration characteristics by using a standard deviation threshold method to obtain an adjusted vibration characteristic; analyzing the vibration change trend of the adjusted vibration characteristics by fast Fourier transform to obtain vibration trend data; Identifying and eliminating noise anomalies in the noise intensity by a standard deviation threshold method to obtain filtered noise; The smoothed temperature, the vibration trend data, and the filtered noise are used as pre-processing data.

3. The intelligent load distribution method for diesel generators according to claim 1, characterized in that: The temperature change rate, vibration frequency peak value and noise decibel mean are extracted from the pre-processed data, and characterization data is obtained through data standardization and characterization processing, including: Extracting the temperature change rate, the vibration frequency peak value, and the noise decibel mean from the preprocessed data, and performing data normalization processing to obtain standardized data, wherein the standardized data includes the standardized temperature change rate, the standardized vibration frequency peak value, and the standardized noise decibel mean; According to the standardized temperature change rate, a time series distribution of the temperature change rate is calculated by a sliding window statistical method and a temperature fluctuation range is determined to obtain temperature distribution data; Performing cluster analysis on the temperature distribution data using the K-means algorithm to obtain a temperature state classification; Calculating a peak value variation of the normalized vibration frequency peak value, and when the peak value variation exceeds a preset peak value variation threshold, marking the corresponding normalized vibration frequency peak value as an abnormal vibration point, thereby obtaining abnormal vibration data; Using a random forest algorithm to perform correlation analysis on the temperature state classification and the abnormal vibration data to obtain a correlation state category; Analyzing the operating state characteristics of the diesel generator by using the Spearman rank correlation coefficient according to the associated state category and the standardized noise decibel mean; The temperature state classification, the abnormal vibration data, the associated state category, and the operating state feature are used as characterization data.

4. The intelligent load distribution method for diesel generators according to claim 1, characterized in that: The training process of the health status assessment model includes: Obtain historical characterization data and corresponding historical health levels; The support vector machine algorithm is used to build the initial health assessment model, the classification boundary is initialized by the RBF kernel function, and the penalty coefficient is set to determine the tolerance error threshold; Based on the historical characterization data, a kernel density estimation method is used to calculate the curvature adjustment parameter of the classification boundary, and the curvature of the classification boundary is dynamically adjusted through the RBF kernel function; According to the historical health level, the penalty coefficient is optimized by a grid search method to determine an error tolerance threshold; Perform cross-validation on the adjusted classification boundaries and calculate the classification accuracy and error distribution set after boundary adjustment; When the error distribution set exceeds the error tolerance threshold, the curvature parameter of the classification boundary is re-optimized using the gradient descent method, and the classification boundary is iteratively updated until the error distribution set does not exceed the error tolerance threshold; The optimized classification boundary and health status level are matched in multi-dimensional space through RBF kernel function mapping to determine the hyperplane equation corresponding to each health level; Combined with the time series characteristics of the historical health levels, the sliding window method is used to extract the dynamic trend parameters of health status changes, and the dynamic trend parameters are correlated with the hyperplane equation to construct the final health status assessment model.

5. The intelligent load distribution method for diesel generators according to claim 1, characterized in that: When the health status level is lower than a preset health level threshold, calculating the load distribution ratio according to the temperature change rate, the vibration frequency peak value, and the noise decibel average value includes: Calculating the relative deviation of the temperature change rate from a preset temperature reference value, and the absolute deviation of the vibration frequency peak from a preset vibration reference value, to obtain an initial deviation data set; Using an entropy weight method to perform weighted processing on the initial deviation data and the noise decibel mean to obtain a weighted deviation sequence; Calculating the load distribution ratio of each generator unit using a preset load distribution ratio formula according to the weighted deviation sequence; The load distribution ratio formula is as follows: Among them, P i represents the load sharing ratio of the i-th generator unit, D i and D j represents the weighted deviation value of the i-th and j-th generator units, which is obtained by weighted calculation of the initial deviation data and the noise decibel mean through the entropy weight method, N D Represents the number of data in the weighted deviation sequence, and ∈ represents the preset proportional error coefficient.

6. The intelligent load distribution method for diesel generators according to claim 1, characterized in that: The method of obtaining historical load demands and determining future load demands using a long short-term memory network algorithm includes: Obtaining historical load demands, and extracting periodic characteristics of the historical load demands using a time series decomposition algorithm; Constructing a long short-term memory network model, setting a preset number of hidden layer units and time step, inputting the periodic features into the long short-term memory network model for training, and outputting a load demand change trend; Based on the load demand change trend and the historical load demand fluctuation range statistics, the quantile regression algorithm is used to calculate the forecast range of future load demand; The Holt-Winters triple exponential smoothing method is used to extract long-term change characteristics of the forecast interval; The long-term variation characteristics are matched with the spatiotemporal characteristics of the periodic characteristics by a dynamic time warping algorithm, and a weighted fusion method is used to generate future load demands.

7. The intelligent load distribution method for a diesel generator according to claim 1, characterized in that: The method of calculating the parameter adjustment step size using a dynamic programming algorithm based on the load distribution ratio and the future load demand, defining the deviation ratio of the set temperature, vibration and noise through preset decision variables, and determining the final operating parameter configuration includes: Based on the future load demand, calculating a parameter adjustment step size of the generator operating parameters by a dynamic programming algorithm; Calculate the deviation ratio data of the temperature change rate, the vibration frequency peak value, and the noise decibel mean value according to the parameter adjustment step size and the preset decision variable definition rule; Calculating an operating parameter configuration according to the deviation ratio data and the load distribution ratio; The operating parameter configuration is calculated using the following formula: Among them, C i represents the final operating parameter configuration value of the i-th generator unit, P i represents the load distribution ratio of the i-th generator unit, Indicates the total load regulation coefficient of the system, ω k and k∈{T,V,N}, that is, ω T ,ω V and ω N Denotes the deviation weight coefficients of temperature, vibration and noise, which are obtained by the decision variable definition rules. Right now and It represents the deviation ratio of the temperature change rate, vibration frequency peak value, and noise decibel mean of the i-th unit, C base_i Represents the reference value of the rated operating parameters of the i-th generator.

8. An intelligent load distribution system for a diesel generator, characterized in that: include: The data acquisition module is used to collect temperature data, vibration signals and noise intensity during the operation of the diesel generator, store them in the database, and obtain the original operating data; A data preprocessing module, configured to perform data preprocessing on the original operation data to obtain preprocessed data; A data characterization module is used to extract the temperature change rate, vibration frequency peak value and noise decibel mean from the pre-processed data, and obtain characterization data through data standardization and characterization processing; A health assessment module, configured to input the characterization data into a pre-established health status assessment model to determine the health status level of the diesel generator; a load ratio calculation module, configured to calculate a load distribution ratio according to the temperature change rate, the vibration frequency peak value, and the noise decibel mean when the health status level is lower than a preset health level threshold; The future load forecasting module is used to obtain historical load demands and determine future load demands using a long short-term memory network algorithm; An operating parameter calculation module is used to calculate the parameter adjustment step size using a dynamic programming algorithm based on the load distribution ratio and the future load demand, define the deviation ratio of the set temperature, vibration and noise through preset decision variables, and determine the final operating parameter configuration.

9. An electronic device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the method for intelligent load distribution of a diesel generator according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the intelligent load distribution method for a diesel generator according to any one of claims 1 to 7.

Images