Intelligent speed regulation universal generator system

Abstract

The invention relates to the technical field of generators, and provides an intelligent speed regulation universal generator system, which is characterized in that a global sensing module performs fusion processing on static characteristic parameters and dynamic characteristic parameters of a prime mover and parameter data of a universal generator, and realizes unified representation of the state of the whole system in combination with sensing state diagram construction; the working condition prediction module realizes short-term, medium-term and long-term layered trend prediction based on a multi-branch structure of a long-short-term memory network and a graph attention network, and ensures that a speed regulation strategy adapts to working condition evolution in advance; the target optimization module constructs operation constraint conditions based on static characteristic parameters of a prime mover, generates an elastic constraint boundary through Bayesian reasoning, and effectively improves the operation efficiency by combining dynamic adjustment of a working condition evolution trend; the execution scheduling module verifies instruction time sequence matching through dynamic time warping, and adjusts execution gain in combination with dynamic characteristic parameters of a prime motor and a load change rate, so that the self-adaptive operation capability of a generator system under complex working conditions is remarkably enhanced.

Description

Technical Field

[0001] The present application relates to the technical field of generators, and in particular to an intelligent speed-regulating universal generator system. Background Art

[0002] With the increasing demand for power supply stability in industrial production and people's livelihood, the speed regulation technology of general generators, as the core equipment for emergency power supply or off-grid power supply, has evolved from mechanical speed regulation to electronic speed regulation. Early mechanical speed regulation relied on centrifugal speed regulators, which adjusted the fuel supply of the prime mover through mechanical feedback to achieve roughly stable speed. Subsequent electronic speed regulation systems introduced sensors and PID control algorithms, which adjusted the actuator action by real-time monitoring of speed deviations, and the speed regulation accuracy was improved to a certain extent.

[0003] In recent years, with the increasing integration of new energy sources and complex load scenarios, generator systems have gradually evolved toward intelligentization. They have begun integrating data collection with simple prediction algorithms, attempting to optimize speed regulation parameters based on historical data. However, traditional systems fragment their processing of prime mover parameters, resulting in a one-sided description of the coordinated operating state of the prime mover and generator. Furthermore, they fail to consider spatial correlations between data, making it difficult to cope with complex operating conditions such as sudden load changes and environmental fluctuations, resulting in suboptimal prediction results. Furthermore, during the target optimization process, the prime mover's operating constraints are often fixed and not dynamically adjusted based on operating trends. This leads to over-constraints or under-constraints during load fluctuations, impacting operational efficiency and equipment safety. Finally, closed-loop execution feedback is also lacking. Most systems fail to address the issues of collaborative perception, dynamic constraint adaptation, and speed regulation execution feedback correction between the prime mover and general-purpose generator under complex operating conditions. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present application provides an intelligent speed-regulating universal generator system, which includes: a global perception module, an operating condition prediction module, a target optimization module and an execution scheduling module;

[0005] The global perception module is used to receive a trigger signal for the target prime mover to be connected to the universal generator, acquire and fuse the static and dynamic characteristic parameters of the target prime mover and the parameter data of the universal generator in real time to form fused power generation data, align the fused power generation data with the historical power generation data in time series, and construct a perception state diagram;

[0006] The operating condition prediction module is used to process the changing trend of the integrated power generation data through the long short-term memory network to deduce the operating condition evolution trend;

[0007] The target optimization module is used to receive the static characteristic parameters of the target prime mover to determine the operating constraints of the target prime mover, analyze the optimization target preset by the user, and generate an instruction sequence including a speed trajectory and actuator action based on the perception state diagram and the working condition evolution trend through reinforcement learning;

[0008] The execution scheduling module is used to receive and verify the instruction sequence, dynamically adjust the execution gain of the instruction sequence according to the load change rate of the general generator and the dynamic characteristic parameters of the target prime mover, and monitor the execution status of the instruction sequence to feedback and update the perception state diagram.

[0009] As an optional implementation, the construction logic of the perception state diagram includes:

[0010] Align the fused power generation data with the historical power generation data, and perform non-uniform time series interpolation on the fused power generation data through dynamic time warping;

[0011] The fused power generation data is divided into adaptive time windows to extract local state features. The local state features in each time window constitute the nodes of the perception state graph.

[0012] The cosine similarity and dynamic time warping distance of adjacent nodes are calculated to determine the edges and connection weights of adjacent nodes, and the directed edges of non-adjacent nodes are added through a causal inference algorithm to construct a perception state graph.

[0013] As an optional implementation, the formation sub-logic of the fused power generation data includes:

[0014] Perform feature anchoring on the static characteristic parameters of the target prime mover, and perform multimodal filtering on the dynamic characteristic parameters of the target prime mover and the parameter data of the general generator;

[0015] Determine the data type and aggregate the correlation of data of the same type. At the same time, use heterogeneous graph neural networks to integrate different types of data across domains, and use the static characteristic parameters of the target prime mover as the initial node features of the heterogeneous graph neural network.

[0016] The dynamic characteristic parameters of the target prime mover and the parameter data of the general generator are used as edge features, and the interaction weights of different types of data are learned through the attention mechanism to form fused power generation data.

[0017] As an optional implementation manner, the deduction logic of the operating condition evolution trend includes:

[0018] The fused power generation data is split into high-frequency components and low-frequency components according to the time dimension. At the same time, the correlation between the fused power generation data is sorted out according to the spatial dimension. The fused power generation data is calculated through mutual information entropy to screen out strongly correlated data pairs.

[0019] The long short-term memory network includes a short-term branch, a medium-term branch, and a long-term branch. The high-frequency component is input into the short-term branch to output a short-term prediction sequence, the low-frequency component and the short-term prediction sequence are input into the medium-term branch to output a stage prediction sequence, and the strongly correlated data pair and the stage prediction sequence are input into the long-term branch to output a trend prediction sequence.

[0020] Dynamically adjust branch weights according to the complexity of the working condition, form a preliminary trend sequence after fusion, receive the perception state graph, calculate the correlation strength between nodes through the graph attention network, generate a topology vector, and fuse the preliminary trend sequence and the topology vector to generate a trend matrix;

[0021] The trend matrix and historical prediction errors are processed through Bayesian inference, and the confidence interval of each prediction result is calculated. At the same time, the confidence interval of the prediction result is corrected according to the degree centrality of the nodes in the perception state diagram to deduce the working condition evolution trend. The working condition evolution trend is then subtracted from the actual data to determine whether to adjust the branch weights of the long and short-term memory network.

[0022] As an optional implementation, the instruction sequence generation logic includes:

[0023] The local state features and edge connection weights in the perception state graph are combined to generate the initial state vector, which is then fused into the state vector after being weighted by the working condition evolution trend through the attention mechanism.

[0024] Parse the user's preset optimization objectives and generate basic rewards, and impose negative rewards on actions that violate operational constraints to determine the reward function for reinforcement learning;

[0025] Based on the state vector and reward function, candidate speed trajectories are generated through Monte Carlo tree search. The candidate speed trajectories are verified by operating constraints to determine the speed trajectory, and actuator actions are generated based on the speed trajectory through the proximal strategy algorithm to generate an instruction sequence.

[0026] As an optional implementation manner, the sub-logic for determining the operation constraint condition includes:

[0027] Receive static characteristic parameters of the target prime mover, set a multi-dimensional constraint space based on rated power, efficiency curve, and structural parameters, and generate initial constraint conditions based on the type of the target prime mover;

[0028] Set elastic constraint boundaries for each initial constraint condition based on historical operating data, and calculate the confidence level of the elastic constraint boundaries through Bayesian reasoning;

[0029] The elastic constraint boundary is dynamically adjusted according to the evolution trend of the operating conditions to determine the operating constraint conditions of the target prime mover.

[0030] As an optional implementation, the feedback update logic of the perception state diagram includes:

[0031] Obtain parameter data of the universal generator during the execution of the instruction sequence, and compare the parameter data with the local state features in the perception state diagram to calculate the feature deviation;

[0032] Configure a deviation threshold. When the feature deviation is less than the deviation threshold, replace the node that perceives the local state feature in the state graph, and update the connection weight of the edge between nodes according to the execution status of the instruction sequence.

[0033] When the feature deviation is greater than or equal to the deviation threshold, the cause of the deviation is analyzed by combining the actuator action and the load change rate of the general generator to determine whether to add a new directed edge to the perception state graph;

[0034] The updated node and edge connection weights are time-series aligned with historical power generation data to verify the rationality of the perception state graph update.

[0035] As an optional implementation, the syndrome logic of the instruction sequence includes:

[0036] Receive the command sequence, extract the timestamps of the actuator actions and the time nodes of the speed trajectory, and verify the timing matching of the actuator actions and the speed trajectory through dynamic time warping;

[0037] Call the elastic constraint boundary in the operating constraint condition and compare the actuator action and speed trajectory with the elastic constraint boundary to determine the cause of the limit violation;

[0038] Combined with the dynamic characteristic parameters of the target prime mover, the execution of the instruction sequence is simulated, and the mechanical state of the universal generator is monitored during the simulation to determine whether to correct the instruction sequence.

[0039] As an optional implementation manner, the gain adjustment sub-logic includes:

[0040] Obtain the load change rate of the general generator in real time, and correlate the load change rate with the speed fluctuation of the target prime mover to generate a load-speed correlation map;

[0041] Based on the load-speed correlation map and the torque response delay of the target prime mover, an execution gain of the instruction sequence is determined, and the instruction sequence is adjusted based on the execution gain;

[0042] Continuously monitor the execution status of the instruction sequence, and subtract the actual execution effect from the expected execution effect to determine whether to adjust the execution gain of the instruction sequence again in combination with the vibration frequency.

[0043] As an optional implementation manner, the static characteristic parameters of the target prime mover include rated power, efficiency curve and structural parameters; the dynamic characteristic parameters of the target prime mover include speed fluctuation, torque response delay and vibration frequency; the parameter data of the general generator include electrical parameters, mechanical state, environmental information and load characteristics; the electrical parameters of the general generator include generator speed, voltage harmonic distortion rate and power factor; the mechanical state of the general generator includes bearing temperature and vibration spectrum; the environmental information of the general generator includes ambient temperature and ambient humidity; and the load characteristics of the general generator include load change rate and power ratio timing.

[0044] Compared with the existing technology, the beneficial effects of the present application are: through the fusion processing of the static characteristic parameters, dynamic characteristic parameters of the prime mover and the parameter data of the general generator by the global perception module, combined with the construction of the perception state diagram, a unified representation of the state of the entire system is achieved, which greatly improves the integrity of the state description and provides comprehensive data support for subsequent optimization decisions; the operating condition prediction module is based on the multi-branch structure of the long short-term memory network, combined with the graph attention network to achieve hierarchical prediction of short-term instantaneous fluctuations, medium-term trend changes and long-term evolution laws, and the confidence interval is corrected through Bayesian inference, which significantly improves the prediction accuracy and ensures that the speed regulation strategy adapts to the operating condition evolution in advance.

[0045] The target optimization module constructs a multi-dimensional constraint space based on the static characteristic parameters of the prime mover, generates elastic constraint boundaries through Bayesian reasoning, and dynamically adjusts them in combination with the evolution trend of working conditions. In the reinforcement learning process, the reward function incorporates both optimization objectives and constraint violation penalties. The generated instruction sequence effectively improves operating efficiency while ensuring equipment safety, and resolves the contradiction between efficiency and safety under fixed constraints; the execution scheduling module verifies the instruction timing matching through dynamic time regularization, adjusts the execution gain in combination with the dynamic characteristic parameters of the prime mover and the load change rate, and establishes a real-time update mechanism for the perception state diagram, which speeds up the response speed of the generator system to load mutations, reduces equipment losses caused by abnormal vibration frequencies, and significantly enhances the adaptive operation capability of the generator system under complex working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be derived from these drawings without inventive work. Among them:

[0047] Figure 1 This is a system flow chart of the intelligent speed-regulating universal generator system provided in an embodiment of the present application;

[0048] Figure 2A logic diagram for deducing the evolution trend of the operating condition of the intelligent speed-regulating universal generator system provided in an embodiment of the present application;

[0049] Figure 3 A sub-logic diagram for determining the operating constraints of the intelligent speed-regulating universal generator system provided in an embodiment of the present application;

[0050] Figure 4 This is a feedback update logic diagram of the perception state diagram of the intelligent speed regulation universal generator system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0051] In order to make the objectives, technical solutions and advantages of the embodiments of the present application more obvious and easy to understand, the technical solutions in the embodiments of the present application are clearly and completely described below in conjunction with the drawings in the specification. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments.

[0052] Example

[0053] like Figure 1 As shown, a system flow chart of an intelligent speed-regulating universal generator system is provided for an embodiment of the present application. The system includes a global perception module, an operating condition prediction module, a target optimization module and an execution scheduling module.

[0054] The global perception module is used to receive the trigger signal of the target prime mover connecting to the general generator, obtain and fuse the static and dynamic characteristic parameters of the target prime mover and the parameter data of the general generator in real time, form fused power generation data, align the fused power generation data with the historical power generation data in time series, and construct a perception state diagram.

[0055] The static characteristic parameters of the target prime mover include rated power, efficiency curve and structural parameters; the dynamic characteristic parameters of the target prime mover include speed fluctuation, torque response delay and vibration frequency; the parameter data of the general generator include electrical parameters, mechanical status, environmental information and load characteristics; the electrical parameters of the general generator include generator speed, voltage harmonic distortion rate and power factor; the mechanical status of the general generator includes bearing temperature and vibration spectrum; the environmental information of the general generator includes ambient temperature and ambient humidity; the load characteristics of the general generator include load change rate and power ratio timing.

[0056] It should be explained that the rated power refers to the maximum power that the prime mover can continuously output under the designed standard operating conditions, and is the basic indicator for measuring the prime mover's work capacity. The standard operating conditions refer to the standard ambient temperature and rated speed. The efficiency curve is used to reflect the efficiency change relationship of the prime mover under different load rates, usually showing a trend of being high in the middle and low at both ends. The load rate refers to the ratio of the actual output power to the rated power. The structural parameters cover the mechanical construction characteristics of the prime mover, including the number of cylinders, stroke length, rotor diameter and bearing model, which directly affect the mechanical strength, heat dissipation capacity and operating stability of the prime mover. The speed fluctuation refers to the fluctuation amplitude of the actual speed of the prime mover from the rated speed when the load changes. The smaller the fluctuation, the stronger the prime mover's adaptability to load changes. The torque response delay refers to the time difference when the prime mover's output torque adjusts from the current value to the target value when the load demand changes. The smaller the delay, the more sensitive the dynamic response of the prime mover. The vibration frequency refers to the frequency component of the vibration signal generated by the mechanical movement when the prime mover is running. Different faults will cause the vibration of a specific frequency to increase.

[0057] Generator speed refers to the rotational speed of the generator rotor, which directly affects the frequency of the output voltage. Voltage harmonic distortion refers to the ratio of the harmonic component to the fundamental component in the output voltage, reflecting the degree of voltage waveform distortion. Excessive distortion can shorten the life of electrical equipment. Power factor represents the ratio of active power to apparent power, measuring energy efficiency. A too low value increases line losses and generator burden. Bearing temperature refers to the temperature of the generator bearing during operation and is a key indicator for determining bearing wear and lubrication status. Exceeding the threshold can cause bearing burnout. Vibration spectrum represents the frequency distribution of the vibration signal during generator operation, including operating status information for components such as the rotor and stator, which can be used for early fault diagnosis. Ambient temperature refers to the air temperature in the generator environment, affecting heat dissipation efficiency and equipment insulation performance. Ambient humidity refers to the water vapor content in the ambient air. Excessive humidity can cause short circuits in electrical components, while excessive humidity can cause electrostatic interference. Load change rate represents the magnitude of load power change per unit time, reflecting load volatility. Sudden load change rates are large, while steady load change rates are small. Power ratio timing refers to the temporal variation of the power share of different load types, affecting the generator's reactive power demand.

[0058] Furthermore, the formation sub-logic of integrated power generation data includes:

[0059] Perform feature anchoring on the static characteristic parameters of the target prime mover, and perform multimodal filtering on the dynamic characteristic parameters of the target prime mover and the parameter data of the general generator;

[0060] Determine the data type and aggregate the correlation of data of the same type. At the same time, use heterogeneous graph neural networks to integrate different types of data across domains, and use the static characteristic parameters of the target prime mover as the initial node features of the heterogeneous graph neural network.

[0061] The dynamic characteristic parameters of the target prime mover and the parameter data of the general generator are used as edge features, and the interaction weights of different types of data are learned through the attention mechanism to form fused power generation data.

[0062] The static characteristic parameters of the target prime movers come from different models of equipment, with different parameter formats and dimensions. Direct use will affect the accuracy of data fusion. Dynamic characteristic parameters and parameter data of general generators are easily affected by noise during the acquisition process and require filtering to ensure data quality. For static characteristic parameters, a parameter standardization mapping library is constructed, and parameters such as the rated power and efficiency curves of different models of prime movers are converted into feature vectors of uniform dimension according to preset rules. For example, discrete points of the efficiency curve are converted into vectors of fixed length through interpolation to complete feature anchoring. Dynamic characteristic parameters and parameter data of general generators are processed through multimodal filtering. Wavelet noise reduction is used for high-frequency signals such as vibration frequency, Kalman filtering is used to eliminate measurement noise for electrical parameters such as voltage harmonic distortion rate, and sliding average filtering is used for slowly changing parameters such as ambient temperature and humidity. Through feature anchoring, the format differences of static parameters of different models of prime movers are eliminated, providing a unified data foundation for subsequent cross-domain fusion. Multimodal filtering processes the characteristics of different types of data, effectively reducing the interference of noise on the data and improving data reliability.

[0063] Data types are diverse, and there are inherent correlations between data of the same type. Aggregation can enhance feature expression. Different types of data also influence each other, that is, ambient temperature affects mechanical state. Cross-domain fusion can explore potential correlations and improve data comprehensiveness. The filtered dynamic characteristic parameters and general generator parameter data are divided into mechanical, electrical, environmental, and load types. For data of the same type, the correlation between the data is calculated using a correlation analysis algorithm, and data with high correlation is aggregated. For example, the bearing temperature and vibration spectrum in the mechanical state are aggregated into a mechanical health feature vector. For cross-domain fusion, a heterogeneous graph neural network is constructed, and the feature vectors corresponding to the static characteristic parameters are used as the initial node features. The dynamic characteristic parameters of different types of target prime movers and the parameter data of general generators are input into the heterogeneous graph neural network as edge features. The interactive relationships between different types of data are learned through inter-layer propagation of the heterogeneous graph neural network. The correlation aggregation of data of the same type reduces data dimensionality and highlights key features. The cross-domain fusion of the heterogeneous graph neural network fully explores the potential correlations between different types of data, forming more comprehensive and representative data features, providing rich information for the subsequent construction of the perception state diagram.

[0064] Different types of data have different degrees of influence on the state of the generator system. For example, changes in load characteristics have a greater impact on the generator system than changes in environmental information. Different weights need to be assigned to highlight the role of key data. In the heterogeneous graph neural network, dynamic characteristic parameters and various parameter data of general generators are used as edge features. The attention mechanism is used to analyze the contribution of different types of data to the state changes of the general generator system in historical operation, and automatically learn and assign interaction weights. For example, when the load change rate is large, the weight of the load feature in the edge feature is increased, and then the weighted edge feature is fused with the initial node feature to form the final fused power generation data. The attention mechanism can adaptively assign weights to different types of data, so that the fused power generation data can better reflect the key state changes of the system and improve the information value of the data. The fused power generation data is the basis for constructing the perception state diagram, which directly affects the accuracy of the state diagram. High-quality fused data can enable subsequent perception state diagrams to more truly reflect the operating status of the generator system.

[0065] Specifically, the construction logic of the perception state diagram includes:

[0066] Align the fused power generation data with the historical power generation data, and perform non-uniform time series interpolation on the fused power generation data through dynamic time warping;

[0067] The fused power generation data is divided into adaptive time windows to extract local state features. The local state features in each time window constitute the nodes of the perception state graph.

[0068] The cosine similarity and dynamic time warping distance of adjacent nodes are calculated to determine the edges and connection weights of adjacent nodes, and the directed edges of non-adjacent nodes are added through a causal inference algorithm to construct a perception state graph.

[0069] The acquisition time intervals of the fused power generation data and the historical power generation data are different, resulting in a time series misalignment problem, which affects the comparison and analysis of the data. At the same time, the fused power generation data has uneven sampling, and interpolation processing is required to ensure the continuity of the data. Based on the time axis of the generator system operation, the timestamps of the fused power generation data and the historical power generation data are matched. For data with mismatched time series, the time series of the fused power generation data is adjusted through the dynamic time warping algorithm to achieve time series alignment between the two. For the non-uniform sampling points in the fused power generation data, dynamic time warping is used for interpolation. According to the changing trends of adjacent sampling points, interpolation data of the intermediate time is generated to keep the fused power generation data uniformly distributed in the time dimension. Time series alignment solves the time misalignment problem of data from different sources, facilitating the comparative analysis of historical power generation data and current fused power generation data. Non-uniform time series interpolation ensures data continuity and provides a coherent data basis for subsequent local state feature extraction.

[0070] The operating state of the generator system changes over time, and the state characteristics of different time segments are different. Fixed time window division cannot accurately capture key state changes. Adaptive time window division can adjust the window length according to the severity of the state change; when the data change rate in the fused power generation data is greater than the preset threshold, the length of the time window is automatically shortened to more finely capture the rapidly changing state, and when the data change rate is less than the preset threshold, the length of the time window is extended to reduce the amount of data processing, and local state characteristics such as the mean, change trend and characteristic peak of the data are extracted in each time window. Each local state feature is used as a node in the perception state diagram; adaptive time window division can flexibly adjust the window length according to the state changes of the generator system, which not only ensures the fine capture of the rapidly changing state, but also improves the data processing efficiency in the stable state. The extracted local state features are used as nodes to provide basic building blocks for the perception state diagram.

[0071] There is a certain correlation between the local state features of adjacent time windows. This correlation can be quantified by similarity and distance calculations and used as the connection weight of the edge. In the case of indirect causal relationships between non-adjacent nodes, causal inference mining is needed to improve the structure of the perception state graph. The cosine similarity of adjacent nodes is calculated to measure the similarity between local state features, and the dynamic time warping distance is calculated to measure the matching degree of time series. The two are combined as the connection weight of the adjacent node edges. For non-adjacent nodes, the causal relationship of data changes in the time windows corresponding to different nodes is analyzed by the causal inference algorithm. When there is a significant causal correlation, a directed edge is added and weighted according to the degree of causal influence. The connection weight of the adjacent node edge accurately reflects the correlation degree of local state features that are continuous in time. The directed edges of non-adjacent nodes mine potential causal relationships, so that the perception state graph includes not only continuous correlations in time, but also discontinuous causal correlations, which more comprehensively reflects the relationship between the states of the generator system.

[0072] A structured approach is needed to integrate the local state characteristics and their correlations of the generator system to intuitively and comprehensively reflect the overall operating status of the generator system; the extracted local state characteristics are used as nodes, and the edges and connection weights of adjacent nodes and the directed edges of non-adjacent nodes are organized according to their correlations to form a complete perception state graph, which includes the attribute information of the nodes and the attribute information of the edges, where the attribute information of the nodes includes the local state characteristics, and the attribute information of the edges includes the connection weights and causal relationships; the perception state graph integrates the state characteristics and correlations of the generator system in a graphical manner, making the operating status of the complex power generation system intuitive and easy to understand, which is convenient for subsequent modules to analyze and process. The perception state graph is an important basis for the operating condition prediction module to process the changing trend of the integrated power generation data and deduce the evolution trend of the operating condition. A high-quality perception state graph can significantly improve the accuracy of the operating condition prediction.

[0073] The operating condition prediction module is used to process the changing trend of the integrated power generation data through the long-short-term memory network to deduce the operating condition evolution trend.

[0074] Specifically, if Figure 2 As shown in Figure 2, the deduction logic of the working condition evolution trend includes:

[0075] The fused power generation data is split into high-frequency components and low-frequency components according to the time dimension. At the same time, the correlation between the fused power generation data is sorted out according to the spatial dimension. The fused power generation data is calculated through mutual information entropy to screen out strongly correlated data pairs.

[0076] The long short-term memory network includes a short-term branch, a medium-term branch, and a long-term branch. The high-frequency component is input into the short-term branch to output a short-term prediction sequence, the low-frequency component and the short-term prediction sequence are input into the medium-term branch to output a stage prediction sequence, and the strongly correlated data pair and the stage prediction sequence are input into the long-term branch to output a trend prediction sequence.

[0077] Dynamically adjust branch weights according to the complexity of the working condition, form a preliminary trend sequence after fusion, receive the perception state graph, calculate the correlation strength between nodes through the graph attention network, generate a topology vector, and fuse the preliminary trend sequence and the topology vector to generate a trend matrix;

[0078] The trend matrix and historical prediction errors are processed through Bayesian inference, and the confidence interval of each prediction result is calculated. At the same time, the confidence interval of the prediction result is corrected according to the degree centrality of the nodes in the perception state diagram to deduce the working condition evolution trend. The working condition evolution trend is then subtracted from the actual data to determine whether to adjust the branch weights of the long and short-term memory network.

[0079] The fusion power generation data contains various data, the changes of which are fast or slow. If they are processed together, the slow trend will be covered by the fast fluctuation. Meanwhile, some data are closely related, and some are loosely related. Therefore, closely related parameter pairs should be selected to avoid irrelevant information interference. In the time dimension, the fusion power generation data is divided into two parts through wavelet transform. The high-frequency component corresponds to the fast-changing data, including the load change rate, which retains its sudden change characteristics. The low-frequency component corresponds to the slow-changing parameters, including the ambient temperature and bearing temperature, which extracts the long-term trend characteristics. In the spatial dimension, the correlation degree of any two data is calculated through mutual information entropy, for example, to determine the correlation between the speed fluctuation and the torque response delay, and the correlation between the load change rate and the power factor. The strongest correlation data pairs are selected, such as the load change rate and the generator speed, and the bearing temperature and the vibration spectrum, and it is marked whether they are causal or concomitant. Time splitting allows the characteristics of fast and slow changes to be retained without interfering with each other. Spatial screening selects closely related data pairs to reduce the influence of irrelevant parameters, so that the long and short term memory network can focus on the parameter relationship that plays a key role in the working condition change, and the input data is more targeted.

[0080] The reasons for working condition changes are different in different time periods. In a short time, it is mainly affected by sudden load changes and instantaneous disturbances. In a medium time, it is affected by environmental changes and accumulated heat of equipment. In a long time, it is affected by equipment aging and wear and tear, and needs to be processed in different branches. Meanwhile, the importance of each branch is different in different working conditions. The short-term branch is more important during sudden loading, while the long-term branch is more important during stable operation. Therefore, the weights of each branch need to be adjusted. The short-term branch receives high-frequency components and uses a network structure with a forgetting gate. When the load changes dramatically, the weight of the forgetting gate is adjusted to make the long and short term memory network pay more attention to the latest high-frequency data. The short-term working condition prediction result is output, and the short-term prediction sequence is obtained, including the peak value of sudden speed change and the time of sudden power change. The medium-term branch receives low-frequency components and the output of the short-term branch, and uses a bidirectional network structure to learn the historical slow trend and predict the future change direction. Combined with the changes of ambient temperature and bearing temperature, the stage working condition prediction result in a medium time is output, and the stage prediction sequence is obtained, including the trend of temperature change and the gradual change of efficiency. The long-term branch receives closely related parameter pairs and the output of the medium-term branch, and refers to similar working condition change cases in history to correct the prediction result using the change path of these cases. The long-term trend prediction result in a long time is output, and the trend prediction sequence is obtained, including the node of equipment aging warning and the time when the efficiency decreases to the critical value.

[0081] Dynamic weight adjustment is performed by an evaluator for operating condition complexity. The evaluator adjusts weights based on the severity of data fluctuations, which takes into account the proportion of rapidly changing data and the number of sudden data changes. When the operating conditions are complex, i.e., sudden load changes or multiple parameters fluctuate simultaneously, the weight of the short-term branch is increased. When the operating conditions are stable, the weight of the long-term branch is increased. After adjustment, the prediction results of each branch are weighted and merged to form a preliminary trend sequence, accompanied by a credibility score for each branch. The multi-branch structure can learn the patterns of operating condition changes in different time periods, making predictions more accurate. Dynamic weight adjustment allows the merged preliminary trend sequence to adapt to the complexity of different operating conditions, focusing on instantaneous changes when the operating conditions are complex and long-term trends when the conditions are stable, thus enhancing adaptability.

[0082] The preliminary trend sequence only reflects the changes in data over time, and does not reflect the spatial correlation and overall structure between data. The node features and edge connection weights in the perception state graph contain the spatial driving relationship of working condition changes. After fusion, the prediction results can be more interpretable. After receiving the perception state graph, the graph attention network is used to calculate the correlation strength between nodes, giving higher attention weights to core nodes with more connections to other nodes, and lower weights to edge nodes with fewer connections. The core nodes include generator speed and load change rate, and the edge nodes include ambient humidity. The weighted results of node features and attention weights are extracted as node importance features. At the same time, the edge connection weights and causal labels are extracted to generate an association strength matrix. The node importance features and the association strength matrix are integrated to form a topological vector.

[0083] The preliminary trend sequence and topological vector are fused through the gated fusion mechanism. When there is a strong causal relationship in the perception state diagram, the proportion of the topological vector in the fusion is increased. When the prediction credibility of the generator speed in the preliminary trend sequence is very high, the proportion of the preliminary trend sequence is increased. The trend matrix generated after fusion includes time dimension, data dimension and correlation dimension, among which the time dimension includes short time, medium time and long time, the parameter dimension includes electrical parameters and mechanical state, etc., and the correlation dimension includes the causal strength between data; after adding the topological vector, the trend matrix can not only predict the changing trend of the data, but also explain the reasons for the data changes, making the prediction results easier to understand. The gated fusion mechanism adjusts the fusion ratio according to the credibility of the time series prediction and the strength of the spatial correlation to avoid interference from weakly correlated information.

[0084] There is uncertainty in the prediction results of the trend matrix. In new working conditions that have not been encountered, the prediction error is large, while in common working conditions dominated by core nodes, the prediction is more credible. Bayesian inference is needed to quantify this uncertainty and adjust the confidence interval according to the degree centrality of the node. At the same time, the deviation between the prediction result and the actual data needs to be fed back to the long-term and short-term memory network to continuously optimize the learning method of each branch. When Bayesian inference is performed, the confidence interval of each prediction result in the trend matrix is ​​first calculated based on the distribution of historical prediction errors, and then corrected according to the number of connections between the nodes in the perception state diagram. The confidence interval of the core nodes with more connections is reduced, and the confidence interval of the edge nodes with fewer connections is expanded. The corrected confidence interval and the prediction result together constitute the working condition evolution. Changing trend; when adjusting the feedback, compare the deviation between the working condition evolution trend and the actual data. When the deviation is large, if the deviation of the short-term branch is large, increase the importance of the high-frequency component in training. If the deviation of the topological vector fusion is large, adjust the weight of the graph attention network calculation node. After the adjustment, mark the samples with large deviation as abnormal working condition samples and store them in the database for subsequent training of the long short-term memory network; Bayesian inference is combined with the degree centrality of the node to correct the confidence interval of the prediction result, so that the prediction reliability of the core node is more accurate, and the uncertainty of the edge node can also be reasonably reflected. The feedback adjustment mechanism gradually reduces the prediction deviation of the long short-term memory network under complex working conditions through targeted optimization of the training method, making the long short-term memory network more stable.

[0085] The target optimization module is used to receive the static characteristic parameters of the target prime mover to determine the operating constraints of the target prime mover, parse the optimization objectives preset by the user, and generate an instruction sequence including speed trajectory and actuator action according to the perception state diagram and working condition evolution trend through reinforcement learning.

[0086] Furthermore, if Figure 3 As shown, the sub-logic for determining the running constraints includes:

[0087] Receive static characteristic parameters of the target prime mover, set a multi-dimensional constraint space based on rated power, efficiency curve, and structural parameters, and generate initial constraint conditions based on the type of the target prime mover;

[0088] Set elastic constraint boundaries for each initial constraint condition based on historical operating data, and calculate the confidence level of the elastic constraint boundaries through Bayesian reasoning;

[0089] The elastic constraint boundary is dynamically adjusted according to the evolution trend of the operating conditions to determine the operating constraint conditions of the target prime mover.

[0090] The static characteristic parameters of the target prime mover are its inherent properties and determine the basic boundaries of operation. Different types of prime movers have different working principles, so the constraints need to be set specifically. If they are not combined with the type of prime mover, the constraint applicability will be poor. After receiving the rated power, it is converted into the upper and lower limit basic values ​​of the output power. The efficiency curve is divided into a high-efficiency range and a low-efficiency range. The low-efficiency range is used as a constraint restricted area. Information such as material and size in the structural parameters is mapped to the basic thresholds of mechanical stress and heat dissipation capacity. The preset rule library is called based on the type of prime mover. For example, diesel prime movers add additional exhaust temperature constraints, while gas prime movers strengthen combustion efficiency constraints, and the initial constraints are integrated to form the initial constraints. Constraints are set based on the inherent properties of the prime mover to ensure the basic nature and rationality of the operating constraints. Combined with the differentiated treatment of the prime mover type, the initial operating constraints are more in line with the actual operating needs and provide a benchmark for the setting of elastic constraint boundaries. Its accuracy directly affects the accuracy of subsequent boundary adjustments, laying the foundation for the dynamic optimization of operating constraints.

[0091] The initial constraints are fixed values, but there are reasonable fluctuations in actual operation. For example, slight load fluctuations may cause short-term power over-limits. Therefore, an elastic constraint boundary needs to be set to accommodate normal fluctuations. At the same time, the reliability of the elastic constraint boundary needs to be quantified to distinguish reasonable fluctuations from abnormal over-limits. The actual fluctuation range of each initial constraint condition is extracted from historical operating data. For example, the fluctuation range of the actual operating value of the rated power around the base value is set as the elastic constraint boundary. Through Bayesian reasoning, the confidence of the elastic constraint boundary is calculated by combining the number of normal operations within the boundary and the number of abnormal operations outside the boundary in the historical operating data. For example, if a boundary can effectively distinguish between normal and abnormal operations 95% of the time in history, the confidence level is 95%. The elastic constraint boundary provides a buffer space for normal operating fluctuations to avoid the decline in operating efficiency caused by excessive constraints. The confidence calculation makes the reliability of the elastic constraint boundary quantifiable, providing a risk reference for subsequent adjustments.

[0092] The working condition evolution trend reflects the change direction of future data, and the elastic constraint boundary needs to be adjusted in advance to reserve a safety margin according to the working condition evolution trend. If it is fixed, it will lose the constraint effect in the trend change; after receiving the working condition evolution trend, the influence of the working condition evolution trend on each elastic constraint boundary is analyzed, for example, when the working condition evolution trend shows that the environmental temperature will continue to rise, the running constraint boundary related to heat dissipation needs to be tightened, and the elastic constraint boundary is adjusted according to the strength of the working condition evolution trend. When the trend is strong, the boundary adjustment range is large, and when the trend is weak, the adjustment range is small. The adjusted boundary needs to update the confidence degree again through Bayesian inference, for example, after the elastic running boundary is adjusted by the working condition evolution trend, the confidence degree is recalculated combined with the running results under similar historical trends, so as to determine the running constraint condition of the target prime mover; the elastic constraint boundary has forward-looking and can actively adapt to the working condition change trend, avoiding the risk of equipment caused by lagging constraint in the trend change. The updated confidence degree ensures the reliability of the adjusted elastic constraint boundary, and the dynamically adjusted elastic constraint boundary constitutes the final running constraint condition, which provides real-time and accurate constraint basis for the generation of the instruction sequence, and ensures that the instruction sequence is optimized within the safety boundary.

[0093] Specifically, the generation logic of the instruction sequence includes:

[0094] The local state features in the perception state graph and the connection weights of the edges are fused to generate an initial state vector, and the working condition evolution trend is weighted through an attention mechanism to generate a state vector;

[0095] The user's preset optimization target is analyzed and a basic reward is generated, and a negative reward is applied to the action that violates the running constraint condition, so as to determine the reward function of reinforcement learning;

[0096] Based on the state vector and the reward function, a candidate speed trajectory is generated through Monte Carlo tree search, the candidate speed trajectory is checked through the running constraint condition to determine the speed trajectory, and the actuator action is generated based on the speed trajectory through the proximal policy algorithm, so as to generate the instruction sequence.

[0097] The perception state diagram includes the local state and correlation relationship of the current generator system, and the operating condition evolution trend reflects the direction of future changes. The fusion of the two can comprehensively describe the current state and future trend of the generator system, providing complete input information for reinforcement learning. Using only a single information will lead to one-sided learning; extract the local state features of each node in the perception state diagram, combine the connection weights of the edges to calculate the comprehensive influence of the nodes, and integrate to form an initial state vector. The importance of each data in the operating condition evolution trend is analyzed through the attention mechanism. For example, the weight of the load change rate in the operating condition evolution trend is higher than the ambient temperature. The weighted operating condition evolution trend features are fused with the initial state vector to form a state vector including spatiotemporal information; the fused state vector covers the details of the current state and the direction of future trends at the same time, so that reinforcement learning can take into account both current optimization and future adaptation. Its integrity directly affects the calculation of the reward function and the optimization of the strategy, providing an accurate state description for the generation of instruction sequences.

[0098] Reinforcement learning requires a clear reward signal to guide strategy optimization. The reward function must reflect the user's optimization goals and punish behaviors that violate constraints to balance optimization and safety. The user's preset optimization goals are analyzed. If the optimization goal is to maximize efficiency, the real-time efficiency value is converted into a basic reward. The higher the efficiency, the greater the reward. At the same time, for actions that violate the operating constraints, negative rewards are set according to the degree of violation. For example, the negative reward for slight violations is small, and the negative reward for serious violations is large. The absolute value of the negative reward is greater than the positive reward of the same degree to ensure safety priority. Finally, the basic reward and negative reward are integrated to form a reward function. The reward function converts abstract optimization goals and constraint requirements into quantifiable signals to balance optimization performance and operational safety. The reward function is the basis for updating the reinforcement learning strategy. Its rationality determines whether the generated instruction sequence meets user needs and safety requirements, providing a clear guide for the optimization of the instruction sequence.

[0099] The speed trajectory is the core goal of the prime mover's operation and must be generated within the operating constraints. The actuator action is the means to achieve the trajectory and must be precisely matched to the trajectory. The combination of the two forms an executable instruction sequence. Optimizing a single link can lead to poor overall performance. Based on the state vector and reward function, a Monte Carlo tree search is used to simulate different speed change paths to generate multiple candidate speed trajectories. These candidate speed trajectories are then verified using the operating constraints. Trajectories that exceed the operating constraints are eliminated, while feasible speed trajectories are retained and the one with the highest reward is selected as the final speed trajectory. Using this speed trajectory as the target, the proximal strategy algorithm generates actuator actions. The timing and amplitude of the actuator actions must match the key nodes of the speed trajectory. For example, the nodes where the speed increases in the speed trajectory correspond to the start time of the actuator action. The generation and verification of the candidate speed trajectories ensure the feasibility and optimization of the speed trajectory. The matching of the actuator actions with the speed trajectory ensures the executable nature of the instruction sequence, enabling the optimization goal to be achieved through specific actions. The generated instruction sequence is directly input into the execution scheduling module. Its accuracy and optimization determine the effectiveness of the execution scheduling, providing specific operational instructions for the intelligent speed regulation of the generator system.

[0100] The execution scheduling module is used to receive and verify the instruction sequence, dynamically adjust the execution gain of the instruction sequence according to the load change rate of the general generator and the dynamic characteristic parameters of the target prime mover, and monitor the execution status of the instruction sequence to feedback and update the perception state diagram.

[0101] Furthermore, the syndrome logic of the instruction sequence includes:

[0102] Receive the command sequence, extract the timestamps of the actuator actions and the time nodes of the speed trajectory, and verify the timing matching of the actuator actions and the speed trajectory through dynamic time warping;

[0103] Call the elastic constraint boundary in the operating constraint condition and compare the actuator action and speed trajectory with the elastic constraint boundary to determine the cause of the limit violation;

[0104] Combined with the dynamic characteristic parameters of the target prime mover, the execution of the instruction sequence is simulated, and the mechanical state of the universal generator is monitored during the simulation to determine whether to correct the instruction sequence.

[0105] The actuator action and speed trajectory need to be precisely coordinated in time. If the timing of the two is misaligned, the speed regulation effect will deviate from expectations. Dynamic time warping can flexibly match asynchronous time series and is suitable for timing verification under complex working conditions. The timestamps of the actuator action and the key time nodes of the speed trajectory are extracted and input into the dynamic time warping algorithm. The dynamic time warping algorithm finds the best matching path by stretching or compressing the time axis of one of the parties. If the total deviation of the matching path is greater than the preset deviation threshold range, it is marked as a timing mismatch and the time period with the largest deviation is output. In this way, the specific time period of timing mismatch can be accurately identified, the speed regulation problem caused by timing mismatch can be solved, and a specific deviation time period reference can be provided for subsequent corrections. If the timing is seriously mismatched, the time synchronization problem must be corrected first, and then the constraint boundary check is performed, otherwise it will affect the accuracy of the over-limit judgment.

[0106] The actuator action or speed trajectory may exceed the elastic constraint boundary, and the reasons for the over-limit are diverse, requiring targeted analysis for effective correction. Merely judging whether the limit is exceeded without tracing the cause will lead to blind correction measures. Call the elastic constraint boundary and compare the actuator action and speed trajectory numbers with the corresponding elastic constraint boundary. If the actuator action exceeds the limit but the speed trajectory is normal, check whether the actuator parameter settings are incorrect. If both exceed the limit at the same time and the timing matching is low, it is determined to be caused by timing misalignment. If the limit is continuously exceeded when the timing matches, it is attributed to the parameter design problem of the initial instruction sequence. In this way, the cause of the over-limit is accurately determined, blind correction is avoided, and subsequent adjustment measures are ensured to directly address the essence of the problem. The clear cause of the over-limit guides the key monitoring direction of the simulation execution, making the simulation more targeted.

[0107] Even if the timing matches and is within the limit, the instruction sequence will still cause potential damage to the mechanical components. Through simulated execution, such hidden problems can be discovered in advance, avoiding equipment losses in actual operation. The instruction sequence is simulated in combination with the dynamic characteristic parameters of the target prime mover for virtual execution. During the simulation process, the mechanical status is monitored. If the bearing temperature shows an abnormal upward trend or the characteristic frequency corresponding to wear appears in the vibration spectrum, it is determined that the instruction sequence needs to be corrected. In this way, the potential harm of the instruction sequence to the mechanical components can be identified in advance, and the limitations of relying solely on constraint boundary verification can be compensated, ensuring the rationality of the instruction sequence in terms of safety and equipment protection. The simulation results determine whether to return to the target optimization module to correct the instruction sequence. If there is a mechanical status risk, the instruction sequence regeneration process is triggered. If there is no error, the execution gain adjustment link is entered.

[0108] Furthermore, the gain adjustment sub-logic includes:

[0109] Obtain the load change rate of the general generator in real time, and correlate the load change rate with the speed fluctuation of the target prime mover to generate a load-speed correlation map;

[0110] Based on the load-speed correlation map and the torque response delay of the target prime mover, an execution gain of the instruction sequence is determined, and the instruction sequence is adjusted based on the execution gain;

[0111] Continuously monitor the execution status of the instruction sequence, and subtract the actual execution effect from the expected execution effect to determine whether to adjust the execution gain of the instruction sequence again in combination with the vibration frequency.

[0112] The load change rate is a key factor affecting speed stability. The speed fluctuation of the target prime mover determines its ability to cope with load changes. The correlation between the two is the basis for determining the execution gain. Ignoring this correlation will make the gain adjustment lack of specificity. The load change rate and the corresponding speed fluctuation are obtained in real time, and a correlation map of the two is drawn through a correlation analysis tool to obtain a load-speed correlation map. In the load-speed correlation map, nodes are used to represent different load change rates and speed fluctuations, and the thickness of the edge is used to represent the strength of the correlation between the two. This intuitively presents the intrinsic connection between the load change rate and speed fluctuation, providing a data-based correlation basis for the execution gain adjustment, so that the gain setting can match the specific load-speed interaction mode. The load-speed correlation map provides a scenario classification basis for determining the execution gain in combination with the torque response delay. Different correlation modes correspond to different gain adjustment strategies.

[0113] The execution gain determines the execution intensity of the instruction sequence. A prime mover with a long torque response delay requires a higher gain to speed up the response speed, but too high a gain will cause overshoot, which needs to be dynamically balanced in combination with the load-speed correlation map. Based on the load-speed correlation map, for modes with high correlation intensity, the execution gain is adjusted in combination with the torque response delay. If the torque response delay is long, the execution gain is increased to compensate for the lag, and if the torque response delay is short, the gain is reduced to avoid overshoot. The determined execution gain is applied to the instruction sequence to adjust the amplitude and rate of the actuator action; thereby, the execution gain is adapted to both the load change requirements and the dynamic response capability of the prime mover, ensuring the timeliness of speed regulation while avoiding system oscillation caused by improper gain. The instruction sequence after preliminary adjustment enters the actual execution stage, providing basic execution effect data for the subsequent secondary adjustment combined with the vibration frequency.

[0114] In actual execution, even if the gain is adapted to the load and torque characteristics, abnormal vibration caused by mechanical resonance still occurs, and the vibration frequency is a key indicator for monitoring such hidden problems, and the gain needs to be optimized again according to the vibration frequency; the vibration frequency during the execution of the instruction sequence is continuously monitored, if a vibration component close to the inherent frequency of the device appears, and the deviation of the actual execution effect from the expected execution effect exceeds the acceptable range, the execution gain of the corresponding frequency band is reduced, if the vibration is normal but the execution effect has a deviation, the amplitude of the execution gain is fine-tuned instead of the frequency; thereby, through feedback correction of the vibration frequency, the risk of mechanical resonance is eliminated, the stability of the execution effect is further improved, the gain adjustment takes into account the dynamic response and mechanical safety, and finally the adjusted instruction sequence enters the stable execution phase, and its execution state will be used for feedback update of the perception state graph to form a closed-loop optimization.

[0115] Specifically, as shown in Figure 4 the feedback update logic of the perception state graph includes:

[0116] Obtain the parameter data of the general generator during the execution of the instruction sequence, compare the parameter data with the local state characteristics in the perception state graph, and calculate the characteristic deviation;

[0117] Configure a deviation threshold, when the characteristic deviation is less than the deviation threshold, replace the nodes of the local state characteristics in the perception state graph, and update the connection weights between the nodes according to the execution state of the instruction sequence;

[0118] When the characteristic deviation is greater than or equal to the deviation threshold, analyze the deviation reason in combination with the actuator action and the load change rate of the general generator, to determine whether to add a new directed edge in the perception state graph;

[0119] Align the connection weights of the updated nodes and edges with the historical power generation data in time sequence to verify the rationality of the update of the perception state graph.

[0120] After the execution of the instruction sequence, the actual parameter data of the general generator will deviate from the local state characteristics in the perception state graph, and such deviation reflects the accuracy of the description of the current working condition by the perception state graph, which is a prerequisite for updating the perception state graph; the parameter data during the execution is compared with the local state characteristics in the corresponding time window of the perception state graph, the deviation of each parameter data from the local state characteristics is judged, including the numerical deviation and the change trend, and the characteristic deviation is formed; thereby, the difference between the perception state graph and the actual working condition is quantified, a clear trigger signal is provided for the update of the perception state graph, blind update or omission of necessary adjustment is avoided, the characteristic deviation result determines the update mode, and data basis is provided for the subsequent deviation threshold judgment.

[0121] When the feature deviation is small, it means that the perception state graph is accurate as a whole and only the node features need to be updated. When the feature deviation is large, there will be new associations that have not been captured, and new directed edges need to be added to improve the structure of the perception state graph. If the deviation size is not distinguished and updated uniformly, it will lead to inefficiency or structural redundancy. When the feature deviation is less than the deviation threshold, the local state features of the corresponding node are replaced with the actual parameter data, and the connection weights of the edges between nodes are updated according to the execution status. When the feature deviation is greater than or equal to the deviation threshold, the cause of the deviation is analyzed in combination with the actuator action and the load change rate of the general generator. If a new causal relationship is found, a new directed edge is added between the corresponding nodes. This not only ensures the timeliness of the perception state graph, but also expands the ability to capture new associations, so that the perception state graph can dynamically adapt to changes in working conditions. The updated nodes and edges provide new execution status and association information for the historical power generation data time series alignment verification, ensuring the accuracy of the verification basis.

[0122] The updated perception state diagram needs to be compared with the historical power generation data to verify whether it conforms to the long-term operation rules of the system and avoid erroneous updates caused by short-term abnormal data; the connection weights of the updated nodes and edges are time-series aligned with the historical power generation data. By comparing the historical perception state diagrams under similar working conditions, it is determined whether the newly added directional edges have similar associations in the historical power generation data, and whether the local state characteristics of the updated nodes are consistent with the historical trends. If they conform to the historical rules, the update is confirmed to be valid; otherwise, it is marked as pending observation and not included in subsequent decisions; through verification of historical power generation data, invalid updates caused by short-term anomalies are filtered out to ensure that the update of the perception state diagram conforms to the inherent operation rules of the system and improve its reliability as the input of subsequent modules. The verified perception state diagram is fed back to the global perception module as the basis for the next round of perception state diagram construction, forming a closed loop of perception, optimization, execution and feedback.

Claims

1. Intelligent speed regulation universal generator system, characterized by: include: Global perception module, working condition prediction module, target optimization module and execution scheduling module; The global perception module is used to receive a trigger signal for the target prime mover to be connected to the universal generator, acquire and fuse the static and dynamic characteristic parameters of the target prime mover and the parameter data of the universal generator in real time to form fused power generation data, align the fused power generation data with the historical power generation data in time series, and construct a perception state diagram; The operating condition prediction module is used to process the changing trend of the integrated power generation data through the long short-term memory network to deduce the operating condition evolution trend; The target optimization module is used to receive the static characteristic parameters of the target prime mover to determine the operating constraints of the target prime mover, analyze the optimization target preset by the user, and generate an instruction sequence including a speed trajectory and actuator action based on the perception state diagram and the working condition evolution trend through reinforcement learning; The execution scheduling module is used to receive and verify the instruction sequence, dynamically adjust the execution gain of the instruction sequence according to the load change rate of the general generator and the dynamic characteristic parameters of the target prime mover, and monitor the execution status of the instruction sequence to feedback and update the perception state diagram.

2. The intelligent speed-regulating universal generator system according to claim 1, characterized in that: The construction logic of the perception state diagram includes: Align the fused power generation data with the historical power generation data, and perform non-uniform time series interpolation on the fused power generation data through dynamic time warping; The fused power generation data is divided into adaptive time windows to extract local state features. The local state features in each time window constitute the nodes of the perception state graph. The cosine similarity and dynamic time warping distance of adjacent nodes are calculated to determine the edges and connection weights of adjacent nodes, and the directed edges of non-adjacent nodes are added through a causal inference algorithm to construct a perception state graph.

3. The intelligent speed-regulating universal generator system according to claim 2, characterized in that: The formation sub-logic of the fused power generation data includes: Perform feature anchoring on the static characteristic parameters of the target prime mover, and perform multimodal filtering on the dynamic characteristic parameters of the target prime mover and the parameter data of the general generator; Determine the data type and aggregate the correlation of data of the same type. At the same time, use heterogeneous graph neural networks to integrate different types of data across domains, and use the static characteristic parameters of the target prime mover as the initial node features of the heterogeneous graph neural network. The dynamic characteristic parameters of the target prime mover and the parameter data of the general generator are used as edge features, and the interaction weights of different types of data are learned through the attention mechanism to form fused power generation data.

4. The intelligent speed-regulating universal generator system according to claim 3, characterized in that: The deduction logic of the working condition evolution trend includes: The fused power generation data is split into high-frequency components and low-frequency components according to the time dimension. At the same time, the correlation between the fused power generation data is sorted out according to the spatial dimension. The fused power generation data is calculated through mutual information entropy to screen out strongly correlated data pairs. The long short-term memory network includes a short-term branch, a medium-term branch, and a long-term branch. The high-frequency component is input into the short-term branch to output a short-term prediction sequence, the low-frequency component and the short-term prediction sequence are input into the medium-term branch to output a stage prediction sequence, and the strongly correlated data pair and the stage prediction sequence are input into the long-term branch to output a trend prediction sequence. Dynamically adjust branch weights according to the complexity of the working condition, form a preliminary trend sequence after fusion, receive the perception state graph, calculate the correlation strength between nodes through the graph attention network, generate a topology vector, and fuse the preliminary trend sequence and the topology vector to generate a trend matrix; The trend matrix and historical prediction errors are processed through Bayesian inference, and the confidence interval of each prediction result is calculated. At the same time, the confidence interval of the prediction result is corrected according to the degree centrality of the nodes in the perception state diagram to deduce the working condition evolution trend. The working condition evolution trend is then subtracted from the actual data to determine whether to adjust the branch weights of the long and short-term memory network.

5. The intelligent speed-regulating universal generator system according to claim 4, characterized in that: The generation logic of the instruction sequence includes: The local state features and edge connection weights in the perception state graph are combined to generate the initial state vector, which is then fused into the state vector after being weighted by the working condition evolution trend through the attention mechanism. Parse the user's preset optimization objectives and generate basic rewards, and impose negative rewards on actions that violate operational constraints to determine the reward function for reinforcement learning; Based on the state vector and reward function, candidate speed trajectories are generated through Monte Carlo tree search. The candidate speed trajectories are verified by operating constraints to determine the speed trajectory, and actuator actions are generated based on the speed trajectory through the proximal strategy algorithm to generate an instruction sequence.

6. The intelligent speed-regulating universal generator system according to claim 5, characterized in that: The sub-logic for determining the operation constraint conditions includes: Receive static characteristic parameters of the target prime mover, set a multi-dimensional constraint space based on rated power, efficiency curve, and structural parameters, and generate initial constraint conditions based on the type of the target prime mover; Set elastic constraint boundaries for each initial constraint condition based on historical operating data, and calculate the confidence level of the elastic constraint boundaries through Bayesian reasoning; The elastic constraint boundary is dynamically adjusted according to the evolution trend of the operating conditions to determine the operating constraint conditions of the target prime mover.

7. The intelligent speed-regulating universal generator system according to claim 6, characterized in that: The feedback update logic of the perception state diagram includes: Obtain parameter data of the universal generator during the execution of the instruction sequence, and compare the parameter data with the local state features in the perception state diagram to calculate the feature deviation; Configure a deviation threshold. When the feature deviation is less than the deviation threshold, replace the node that perceives the local state feature in the state graph, and update the connection weight of the edge between nodes according to the execution status of the instruction sequence. When the feature deviation is greater than or equal to the deviation threshold, the cause of the deviation is analyzed by combining the actuator action and the load change rate of the general generator to determine whether to add a new directed edge to the perception state graph; The updated node and edge connection weights are time-series aligned with historical power generation data to verify the rationality of the perception state graph update.

8. The intelligent speed-regulating universal generator system according to claim 7, characterized in that: The syndrome logic of the instruction sequence includes: Receive the command sequence, extract the timestamps of the actuator actions and the time nodes of the speed trajectory, and verify the timing matching of the actuator actions and the speed trajectory through dynamic time warping; Call the elastic constraint boundary in the operating constraint condition and compare the actuator action and speed trajectory with the elastic constraint boundary to determine the cause of the limit violation; Combined with the dynamic characteristic parameters of the target prime mover, the execution of the instruction sequence is simulated, and the mechanical state of the universal generator is monitored during the simulation to determine whether to correct the instruction sequence.

9. The intelligent speed-regulating universal generator system according to claim 8, characterized in that: The gain adjustment sub-logic includes: Obtain the load change rate of the general generator in real time, and correlate the load change rate with the speed fluctuation of the target prime mover to generate a load-speed correlation map; Based on the load-speed correlation map and the torque response delay of the target prime mover, an execution gain of the instruction sequence is determined, and the instruction sequence is adjusted based on the execution gain; Continuously monitor the execution status of the instruction sequence, and subtract the actual execution effect from the expected execution effect to determine whether to adjust the execution gain of the instruction sequence again in combination with the vibration frequency.

10. The intelligent speed-regulating universal generator system according to claim 9, characterized in that: The static characteristic parameters of the target prime mover include rated power, efficiency curve and structural parameters; the dynamic characteristic parameters of the target prime mover include speed fluctuation, torque response delay and vibration frequency; the parameter data of the general generator include electrical parameters, mechanical state, environmental information and load characteristics; the electrical parameters of the general generator include generator speed, voltage harmonic distortion rate and power factor; the mechanical state of the general generator includes bearing temperature and vibration spectrum; the environmental information of the general generator includes ambient temperature and ambient humidity; and the load characteristics of the general generator include load change rate and power ratio timing.

Images