Intelligent power plant power rate intelligent regulation method and system

Abstract

The present application relates to the technical field of power plant management, and discloses a kind of intelligent power plant power factor intelligent regulation and control method and system, collect power plant equipment energy consumption data, environmental parameter, load demand data, equipment operating state data and historical power factor data, align multi-source data in combination with DTW algorithm, obtain time series data;GNN network is used to topologically model power plant equipment;Time series data is input into Transform model, and the result output by GNN network is combined to carry out feature extraction, and obtain future 24 hours load prediction result;Based on load prediction result, through the joint application of particle swarm optimization algorithm and DQN network, regulation and control scheme is generated;Regulation and control scheme is converted into regulation and control instruction, regulation and control instruction is issued, and regulation and control data is collected to carry out real-time feedback;The present application realizes the real-time prediction of power factor, and the response time of regulation and control is shortened compared with traditional manual regulation and control, reduces plant power factor, and improves response speed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power plant management, in particular to a smart power plant power rate intelligent regulation method and system. BACKGROUND

[0002] As a major energy consumer, the power rate of a power plant is one of the core indicators for measuring its energy utilization efficiency, and the power rate regulation effect directly affects the operating cost and environmental performance of the power plant. The current power plant power rate regulation mainly has the following shortcomings: the traditional regulation method relies on manual experience, and the response to changes in working conditions is lagging, making it difficult to cope with complex and variable power generation loads, equipment states and environmental parameters and other factors; existing regulation algorithms are mostly based on single factor analysis, without considering the coupling effects of multi-dimensional variables, resulting in low regulation accuracy; at the same time, the existing methods lack self-learning ability, and when the working conditions of the power plant change greatly, the regulation effect decreases significantly. SUMMARY

[0003] The present application relates to the technical field of power plant management, in particular to a smart power plant power rate intelligent regulation method and system.

[0004] The first aspect of the present application provides a smart power plant power rate intelligent regulation method, which comprises the following steps:

[0005] Collecting power plant equipment energy consumption data, environmental parameters, load demand data, equipment operating state data and historical power rate data, aligning multi-source data with a DTW algorithm to obtain time series data;

[0006] Topological modeling of power plant equipment is performed using a GNN network, fixed load equipment and adjustable load equipment are divided, and a attention mechanism is introduced to quantify the sensitivity of power plant equipment to grid fluctuations;

[0007] The time series data is input into a Transformer model, and the results output by the GNN network are used for feature extraction to obtain a 24-hour load prediction result;

[0008] Based on the load prediction result, a regulation scheme is generated by joint application of a particle swarm optimization algorithm and a DQN network;

[0009] The regulation scheme is converted into a regulation instruction, the regulation instruction is issued, and the data after regulation is collected for real-time feedback.

[0010] Optionally, in the first implementation manner of the first aspect of the present application, the collecting of power plant equipment energy consumption data, environmental parameters, load demand data, equipment operating state data and historical power rate data, and the aligning of multi-source data with a DTW algorithm to obtain time series data comprises:

[0011] Taking a minute-level timestamp of device running state data as a reference benchmark, data of different sampling frequencies are converted into time sequence forms, and each data point corresponds to a unique time mark;

[0012] Distances between different data sequences are calculated, an optimal matching path is found through dynamic programming, data with time offset is adjusted, and time sequence data of a unified time scale is obtained.

[0013] Optionally, in the second implementation manner of the first aspect of the present application, the GNN network is used to model the topology of the power plant equipment, and the fixed load equipment and the adjustable load equipment are divided, including:

[0014] The power plant equipment is taken as nodes, and the energy transmission relationship between the equipment is taken as edges, and an initial topology graph is constructed;

[0015] Based on the initial topology graph, the GNN network updates the features of each node by aggregating the neighbor node information, completes the modeling of the device association relationship, and forms a complete power plant equipment topology structure graph;

[0016] The nodes in the power plant equipment topology structure graph are traversed, the load characteristics of each device are analyzed, the fixed load equipment and the adjustable load equipment are divided, and a classification result is obtained;

[0017] The fixed load equipment is a device with stable running state and unable to change energy consumption through regulation and control, and the adjustable load equipment is a device with energy consumption changing with running parameters.

[0018] Optionally, in the third implementation manner of the first aspect of the present application, the attention mechanism is introduced to quantify the sensitivity of the power plant equipment to the grid fluctuation, including:

[0019] In the GNN network, the attention weight of each node to its neighbor node is calculated, and the normalized attention weight is obtained through the LeakyReLU activation function and the Softmax normalization;

[0020] The normalized attention weight is quantified as the device sensitivity, and the device sensitivity matrix is generated according to the attention weight and the classification result, wherein the row of the matrix represents the device, and the list represents the sensitivity index.

[0021] Optionally, in the fourth implementation manner of the first aspect of the present application, the time sequence data is input into the Transformer model, the results output by the GNN network are combined for feature extraction, and the future 24-hour load prediction result is obtained, including:

[0022] The time sequence data is input into the encoder layer of the Transformer model, and the device topology association information output by the GNN network is input into the self-attention mechanism module of the encoder as auxiliary features;

[0023] The encoder first extracts time features based on the time series data, and then distinguishes and extracts the stable features of the fixed load device and the fluctuation features of the adjustable load device, and generates a comprehensive feature vector by fusing time correlation, device topology correlation and load attributes through multi-layer processing;

[0024] The comprehensive feature vector is input into the decoder layer of the Transformer model, and the decoder generates load prediction values for the next 24 hours through cross-attention mechanism layer by layer, and finally outputs the complete load prediction results for the next 24 hours.

[0025] Optionally, in the fifth implementation manner of the first aspect of the present application, the load prediction results are used to generate a regulation scheme by joint application of a particle swarm optimization algorithm and a DQN network, including:

[0026] The future 24-hour load prediction results, the adjustable load device and the device sensitivity are input to initialize the particle swarm, each particle represents a set of regulation parameter combinations, and the fitness is calculated in combination with the device sensitivity;

[0027] The particles update the position and speed by tracking the individual optimal solution and the global optimal solution, move to better regulation parameter combinations, and after 50 iterations, a set of regulation parameter candidate set with high fitness is obtained;

[0028] The state space and the action space of the DQN network are determined, and each regulation parameter combination in the regulation parameter candidate set is scored through the experience replay mechanism;

[0029] The regulation parameter candidate set is screened according to the scoring results, and the regulation parameter combination with the highest score is selected as the final regulation scheme.

[0030] Optionally, in the sixth implementation manner of the first aspect of the present application, the load prediction results, the regulation parameter combinations in the regulation parameter candidate set and the real-time running state of the device are used as the state space;

[0031] The regulation parameter combinations in the regulation parameter candidate set are used as optional actions of the action space, and device association labels are added to each action in combination with the device topology correlation information.

[0032] The second aspect of the present application provides a smart power plant power consumption rate intelligent regulation system, which comprises:

[0033] The acquisition module is used for acquiring power plant device energy consumption data, environmental parameters, load demand data, device running state data and historical power consumption rate data, aligning multi-source data in combination with the DTW algorithm, and obtaining time series data;

[0034] a topological modeling module, configured to perform topological modeling on the power plant equipment by using a GNN network, divide fixed load equipment and adjustable load equipment, and introduce an attention mechanism to quantify the sensitivity of the power plant equipment to power grid fluctuations;

[0035] a feature extraction module, configured to input time series data into a Transformer model, combine the results output by the GNN network to perform feature extraction, and obtain a 24-hour load prediction result;

[0036] a generation module, configured to generate a regulation scheme based on the load prediction result by jointly applying a particle swarm optimization algorithm and a DQN network;

[0037] a regulation module, configured to convert the regulation scheme into regulation instructions, issue the regulation instructions, and collect regulated data for real-time feedback.

[0038] A third aspect of the present application provides a smart power plant electricity rate intelligent regulation device, which comprises a memory and at least one processor, and the memory stores instructions; the at least one processor invokes the instructions in the memory to enable the smart power plant electricity rate intelligent regulation device to perform each step of the smart power plant electricity rate intelligent regulation method according to any one of the above.

[0039] A fourth aspect of the present application provides a computer-readable storage medium, which stores instructions, and the instructions are executed by a processor to implement each step of the smart power plant electricity rate intelligent regulation method according to any one of the above.

[0040] In the technical solution provided by the present application, power plant equipment energy consumption data, environmental parameters, load demand data, equipment operating state data and historical electricity rate data are collected, multi-source data is aligned by using a DTW algorithm to obtain time series data; a GNN network is used to perform topological modeling on the power plant equipment, to divide fixed load equipment and adjustable load equipment, and an attention mechanism is introduced to quantify the sensitivity of the power plant equipment to power grid fluctuations; time series data is input into a Transformer model, and the results output by the GNN network are combined for feature extraction to obtain a 24-hour load prediction result; based on the load prediction result, a regulation scheme is generated by jointly applying a particle swarm optimization algorithm and a DQN network; the regulation scheme is converted into regulation instructions, the regulation instructions are issued, and regulated data is collected for real-time feedback; the present application realizes real-time prediction of electricity rate, shortens the response time of regulation compared with traditional manual regulation, reduces the plant electricity rate, improves the response speed, and avoids the limitations of single-factor regulation. BRIEF DESCRIPTION OF DRAWINGS

[0041] Various other advantages and benefits will become apparent to those of ordinary skill in the art, upon reading the following detailed description of the preferred embodiments. The accompanying drawings are included to provide a description of preferred embodiments, and are not meant to limit the present application.

[0042] Figure 1 A flow chart of the intelligent power plant power rate intelligent regulation method provided by the embodiment of the present application is shown in the figure.

[0043] Figure 2 A structural schematic diagram of the intelligent power plant power rate intelligent regulation system provided by the embodiment of the present application is shown in the figure.

[0044] Figure 3 A structural schematic diagram of the intelligent power plant power rate intelligent regulation device provided by the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0045] The terms "first", "second", "third", "fourth" and the like in the description and claims of the present application, and above drawings (if there is) are used for distinguishing between similar objects, and do not necessarily have to imply a described particular order or sequence. It is to be understood that the data used in this way can be interchanged, where appropriate, so that the embodiments described herein can be carried out in sequences other than those illustrated or otherwise described herein. Moreover, the terms "comprise" or "have" and any variations thereof are intended to cover a non-exclusive inclusion, for example, a process, method, article, or apparatus that comprises a list of steps or units not necessarily limited to those specifically listed, but can include additional steps or units not expressly listed or inherent to such process, method, article, or apparatus.

[0046] For the convenience of understanding, the specific flow of the embodiment of the present application is described below, please refer to Figure 1 The flow chart of the intelligent power plant power rate intelligent regulation method provided by the embodiment of the present application, the method specifically includes the following steps:

[0047] Step 101, collecting power plant equipment energy consumption data, environmental parameters, load demand data, equipment running state data and historical power rate data, aligning multi-source data by combining DTW algorithm to obtain time series data;

[0048] In this embodiment, when collecting energy consumption data from power plant equipment, the data covers core power generation equipment such as boilers, steam turbines, and generators, as well as auxiliary equipment such as circulating water pumps, induced draft fans, and feedwater pumps. Real-time data on hourly coal consumption, electricity consumption, and steam consumption are acquired through the equipment's built-in energy consumption monitoring sensors. Simultaneously, energy consumption data from environmental protection equipment such as desulfurization towers and denitrification devices are recorded. Environmental parameter collection includes real-time temperature, humidity, and air pressure within the plant area, as well as meteorological data such as outdoor wind speed and precipitation. Data is collected every 5 minutes using environmental sensors deployed at different locations within the plant to ensure that the data reflects environmental differences in different areas. Load demand data is primarily obtained through integration with the power grid dispatch system. The system generates hourly power load commands and records the deviation between the actual power generation and the load commands. Equipment operating status data is collected for key operating parameters, such as boiler outlet temperature, pressure, and water level; turbine speed, vibration value, and exhaust steam temperature; and generator voltage, current, and power factor. This data is collected every minute through the real-time monitoring module of the equipment control system. Historical electricity consumption data is extracted from the historical database of the power plant's energy management system, covering daily electricity consumption, monthly average electricity consumption, and seasonal electricity consumption changes over the past three years. This data is also linked to equipment operating records and environmental conditions for the corresponding time periods, forming a complete historical data chain.

[0049] In this embodiment, using the minute-level timestamps of the device operating status data as a reference, data with different sampling frequencies are converted into time series format, with each data point corresponding to a unique time marker; the distance between different data series is calculated, and the optimal matching path is found through dynamic programming; data with time offsets are adjusted to obtain time series data with a unified time scale.

[0050] Step 102: Use a GNN network to perform topology modeling of power plant equipment, divide the equipment into fixed load equipment and adjustable load equipment, and introduce an attention mechanism to quantify the sensitivity of power plant equipment to grid fluctuations;

[0051] In this embodiment, power plant equipment is used as nodes, and the energy transmission relationship between equipment is used as edges to construct an initial topology graph. Based on the initial topology graph, the GNN network updates the features of each node by aggregating neighbor node information, completes the modeling of equipment association relationships, and forms a complete power plant equipment topology graph. The nodes in the power plant equipment topology graph are traversed, the load characteristics of each device are analyzed, and fixed load devices and adjustable load devices are classified to obtain classification results. Among them, fixed load devices are devices with stable operating states and whose energy consumption cannot be changed by adjustment, while adjustable load devices are devices whose energy consumption changes with operating parameters.

[0052] In this embodiment, in the GNN network, the attention weights of each node to its neighboring nodes are calculated, and normalized attention weights are obtained by using the LeakyReLU activation function and Softmax normalization. The normalized attention weights are quantized into device sensitivity, and a device sensitivity matrix is ​​generated based on the attention weights and classification results, where the rows of the matrix represent devices and the columns represent sensitivity indices.

[0053] In this embodiment, all devices involved in energy conversion or transmission, such as boilers, steam turbines, generators, circulating water pumps, and induced draft fans, are treated as individual nodes. Each node is labeled with its device name and basic attributes, such as rated power and device type. Simultaneously, edge connections are determined based on the actual energy flow relationships between the devices. For example, if a boiler transmits steam to a steam turbine, an edge is established between the boiler node and the steam turbine node; if a generator outputs electricity to the grid, an edge is established between the generator node and the grid interface node. The edge labels indicate the energy transmission type, thus forming the initial topology graph.

[0054] The GNN network first assigns an initial feature vector to each node, containing basic device attributes and real-time energy consumption data. Each node first collects the features of all neighboring nodes corresponding to all connected edges, and then combines the neighboring features with its own features through a feature fusion algorithm such as weighted averaging, with the edge weights set according to the energy transmission intensity, to generate a new node feature vector. After multiple iterations, the node features gradually contain complete information about the associated devices, ultimately forming a complete topology graph that includes the device association strength and energy interaction patterns.

[0055] During the traversal, historical load data for each node is extracted, such as hourly energy consumption records for the past 30 days and load impact data in the topology, such as energy consumption changes of the device when other devices are adjusted. The focus is on analyzing the fluctuation range and control responsiveness of the load data, calculating the maximum fluctuation amplitude and standard deviation of the historical load, and examining the energy consumption change rate of the device when adjacent adjustable devices are adjusted.

[0056] Fixed load equipment must meet two conditions: the maximum historical load fluctuation is less than 5%, and its own energy consumption change rate is less than 3% when the adjacent equipment is adjusted to the maximum amplitude; adjustable load equipment is characterized by a historical load fluctuation greater than 8%, or an energy consumption change rate exceeding 10% when the operating parameters are adjusted. Finally, the two types of equipment are classified according to the above standards and the classification results are output.

[0057] In this embodiment, the normalized attention weight obtained through the attention mechanism is directly used as a quantitative indicator of the equipment sensitivity, and the higher the weight value, the stronger the sensitivity of the equipment to the grid fluctuation. At the same time, combined with the equipment classification results output by the GNN network, such as fixed load equipment, adjustable load equipment and equipment type information such as boiler, steam turbine, circulating water pump, etc., the sensitivity indicator dimension is determined, including grid load fluctuation sensitivity, environmental parameter change sensitivity, and adjacent equipment state correlation sensitivity. An initial matrix framework is constructed with equipment as rows and sensitivity indicators as columns, and each item of attention weight corresponding to each equipment is filled into the corresponding row and column position, for example, the attention weight of the steam turbine in the grid load fluctuation dimension is 0.8, which is filled into 0.8 at the intersection of the steam turbine row and the grid load fluctuation sensitivity column. Finally, a complete equipment sensitivity matrix is generated, and each value in the matrix corresponds to the quantitative result of a specific equipment under a specific sensitivity indicator.

[0058] Step 103, input the time series data into the Transformer model, and combine the results output by the GNN network to extract features, to obtain the future 24-hour load prediction result;

[0059] In this embodiment, the time series data is input into the encoder layer of the Transformer model, and the device topology correlation information output by the GNN network is input into the self-attention mechanism module of the encoder as auxiliary features; the encoder first extracts time features based on the time series data, then distinguishes and extracts the stable features of fixed load equipment and the fluctuation features of adjustable load equipment, and generates a comprehensive feature vector that integrates time correlation, device topology correlation and load attributes through multi-layer processing; the comprehensive feature vector is input into the decoder layer of the Transformer model, and the decoder generates the load prediction value of the future 24 hours through cross-attention mechanism layer by layer, and finally outputs the complete 24-hour load prediction result.

[0060] In this embodiment, the time series data includes real-time energy consumption, environmental parameters, historical load, etc. of each equipment in the power plant, which are sorted according to minute-level timestamps, and after being input into the encoder layer, they are first converted into vector form to adapt to the model processing format; the device topology correlation information output by the GNN network includes the connection relationship between devices such as the steam transmission correlation between the boiler and the steam turbine and the correlation strength value. These information is transmitted to the self-attention mechanism module in the form of a feature matrix, providing a reference basis for the correlation between devices for the module, so that the self-attention mechanism can adjust the weight in calculation considering the actual correlation between devices;

[0061] The encoder layer first performs sliding window processing on the time series data to capture the data change law in different time segments and extract time features such as energy consumption rising during the morning peak period. Then, according to the device classification results output by the GNN network, the stable feature energy consumption data of fixed load devices such as lighting devices is extracted separately, and the fluctuation feature energy consumption of adjustable load devices such as boilers is extracted separately. After iterative processing of multiple encoder layers, each layer further fuses the features output by the previous layer with the device topology association information to finally generate a comprehensive feature vector containing time dimension rules, device association relationships, and load type attributes.

[0062] After the comprehensive feature vector is input into the decoder layer, the decoder first constructs an initial prediction framework based on the vector. The cross-attention mechanism focuses on key information related to future load in the comprehensive feature vector, such as the association between the fluctuation feature of adjustable load devices and historical load, the influence degree of certain devices on the overall load in the device topology association, etc. The decoder generates prediction values layer by layer in units of hours. After generating a load prediction value for each hour, the prediction result is incorporated into the prediction reference for the next hour. After 24 times of layer-by-layer generation, the complete load prediction result covering the next 24 hours is finally obtained.

[0063] Step 104, based on the load prediction result, a control scheme is generated by joint application of the particle swarm optimization algorithm and the DQN network.

[0064] In this embodiment, the future 24-hour load prediction result, adjustable load devices, and device sensitivity are input to initialize the particle swarm, and each particle represents a combination of control parameters. The fitness is calculated in combination with the device sensitivity.

[0065] The particles update their positions and velocities by tracking the individual optimal solution and the global optimal solution, and move towards better control parameter combinations. After 50 iterations, a set of control parameter candidates with high fitness is obtained.

[0066] The state space and action space of the DQN network are determined, and each control parameter combination in the control parameter candidate set is scored through the experience replay mechanism.

[0067] According to the scoring results, the control parameter candidate set is screened, and the control parameter combination with the highest score is selected as the final control scheme.

[0068] In this embodiment, the load prediction result, the control parameter combination in the control parameter candidate set, and the real-time operating state of the device are used as the state space. The control parameter combination in the control parameter candidate set is used as the selectable action of the action space, and the device association label is added to each action in combination with the device topology association information.

[0069] The load prediction result includes a predicted load value of each hour within the next 24 hours in a time sequence, and is included in a state space to provide a load change trend reference for regulation and control. The regulation parameter combination in the regulation parameter candidate set covers specific adjustment parameters of each adjustable device, such as a boiler combustion efficiency adjustment range, a steam turbine admission value adjustment value, and the like, which can reflect a possible regulation direction. The device real-time running state includes current key running indicators of each device, such as a real-time outlet temperature and pressure of the boiler, a rotating speed and vibration value of the steam turbine, and the like. By integrating the three aspects of information, a state space that comprehensively reflects a current regulation basis condition is formed.

[0070] Each regulation parameter combination in the regulation parameter candidate set corresponds to a selectable action in the action space, and each action is specifically directed to a specific regulation operation. Meanwhile, according to the device topology association information, a main device involved in the action and an associated device affected by the main device are determined, for example, an action of adjusting a boiler combustion efficiency, which includes a steam turbine and a steam pipeline, and the like. These association relationships are attached to the corresponding action in the form of a label, such as a boiler combustion efficiency adjustment-association steam turbine, steam pipeline, so that the action can clearly reflect a potential impact on the related devices.

[0071] In the embodiment, the future 24-hour load prediction result determines load demand in different time periods, the adjustable load device determines an object range that can be subjected to a regulation operation, and the device sensitivity reflects a response degree of each device to the regulation. When the particle swarm is initialized, a certain number of particles are generated according to the number of adjustable load devices and the allowed range of regulation parameters, and each particle includes specific regulation parameters of each adjustable load device, such as a boiler combustion efficiency adjustment of 3% and a steam turbine admission value adjustment of 2%. When the fitness is calculated, the device sensitivity is mainly referred to, and a higher weight is given to a device with high sensitivity. When the regulation parameter combination can make the deviation between the load prediction result and the target load smaller, the fitness is higher.

[0072] The individual optimal solution is a regulation parameter combination with the highest fitness in the history of each particle itself, and the global optimal solution is a regulation parameter combination with the highest fitness in all particles. The particles adjust the moving direction and speed according to the two optimal solutions. If the fitness of the current particle is lower than the individual optimal solution, the parameters are adjusted in the direction of the individual optimal solution. If the individual optimal solution is lower than the global optimal solution, the parameters are further optimized in the direction of the global optimal solution. After 50 iterations, the particles with low fitness are eliminated, and the particles with high fitness are retained, and finally the regulation parameter candidate set is formed.

[0073] The state space of the DQN network contains the future 24-hour load prediction result, the combination of the regulation parameters in the candidate set of regulation parameters, and the real-time running state of the device, and the action space takes the combination of the regulation parameters in the candidate set of regulation parameters as the selectable action; the experience replay mechanism calls historical regulation data, which records the implementation effects of different combinations of regulation parameters in a similar load scenario in the past; each combination of regulation parameters in the candidate set is compared with the historical data, and the regulation effect of the same or similar combination in the history is referred to, such as whether it can effectively make the load reach the target value, and each combination is scored;

[0074] During the screening process, the combination of regulation parameters with a higher score is preferentially retained as the main basis; if there are multiple combinations with the same or similar scores, the combination with less impact on the associated devices is selected by combining the device topology association information and considering the impact of the regulation operation on the associated devices; finally, the combination of regulation parameters with the highest score and reasonable impact on the associated devices is determined as the final regulation scheme.

[0075] Step 105, convert the regulation scheme into regulation instructions, issue the regulation instructions, and collect the data after regulation for real-time feedback.

[0076] In this embodiment, when converting the regulation scheme into regulation instructions, first analyze the core regulation parameters in the scheme, such as the boiler combustion efficiency adjustment value, the steam turbine admission flow adjustment value, etc., convert these parameters into standardized instructions recognizable by the device according to the format specified by the power plant device control protocol, set exclusive instruction codes for different types of devices, such as the code "GL-XX" for the boiler and the code "QLJ-XX" for the steam turbine, and clearly specify the specific time, adjustment range and target parameter value of the instruction execution, such as GL-01, 10:05 execution, 3% increase in combustion efficiency, target value 92%. Subsequently, issue the regulation instructions through the industrial control network inside the power plant, determine the issuance sequence according to the device topology association relationship, such as issuing the boiler instructions first and then issuing the steam turbine instructions, with an interval of 30 seconds to avoid fluctuations caused by synchronous adjustment of associated devices, and monitor the transmission state in real time during the instruction issuance process, and if transmission interruption occurs, immediately start the standby transmission channel to resend. At the same time, after the execution of the instructions, the real-time running data of the corresponding device is collected through the sensors and data collection terminals of the device, such as the adjusted boiler combustion efficiency, steam turbine admission flow, and overall power consumption rate data, and these data are sorted by timestamp and transmitted to the central control system to form a real-time feedback data stream.

[0077] Please refer to Figure 2 The intelligent power plant power consumption rate intelligent regulation system provided by the embodiment of the present application has the structure as shown in the figure, and the system comprises:

[0078] The collection module is used for collecting power plant equipment energy consumption data, environmental parameters, load demand data, equipment operation state data and historical power consumption rate data, aligning multi-source data by combining a DTW algorithm, and obtaining time series data.

[0079] The topology modeling module is used for adopting a GNN network to perform topology modeling on the power plant equipment, dividing fixed load equipment and adjustable load equipment, and introducing an attention mechanism to quantify the sensitivity of the power plant equipment to power grid fluctuations.

[0080] The feature extraction module is used for inputting the time series data into a Transformer model, performing feature extraction in combination with the output result of the GNN network, and obtaining a 24-hour future load prediction result.

[0081] The generation module is used for generating a regulation and control scheme based on the load prediction result by joint application of a particle swarm optimization algorithm and a DQN network.

[0082] The regulation and control module is used for converting the regulation and control scheme into regulation and control instructions, issuing the regulation and control instructions, and collecting data after regulation and control for real-time feedback.

[0083] Figure 3 The smart power plant power consumption rate intelligent regulation and control device 300 can be greatly different due to different configurations or performances, and can include one or more processors (central processing units, CPUs) 310 (for example, one or more processors) and a memory 320, one or more storage media 330 (for example, one or more mass storage devices) storing application programs 333 or data 332. The memory 320 and the storage medium 330 can be temporary storage or persistent storage. The programs stored in the storage medium 330 can include one or more modules (not shown in the figure), each module can include a series of instruction operations in the smart power plant power consumption rate intelligent regulation and control device 300. Further, the processor 310 can be configured to communicate with the storage medium 330, execute a series of instruction operations in the storage medium 330 on the smart power plant power consumption rate intelligent regulation and control device 300, to realize the method provided by the above embodiment.

[0084] The smart power plant power consumption rate intelligent regulation and control device 300 can also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art can understand that, Figure 3The structure of the intelligent power plant power rate intelligent regulation and control device shown does not constitute a limitation on the computer device provided by the present application, and can include more or fewer components than shown, or combine certain components, or different component arrangements.

[0085] The present application also provides a computer readable storage medium, which can be a non-volatile computer readable storage medium, and can also be a volatile computer readable storage medium, and the computer readable storage medium has instructions stored therein, and when the instructions are run on a computer, the computer executes the steps of the intelligent power plant power rate intelligent regulation and control method provided by the above embodiments.

[0086] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described device or apparatus, unit can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0087] The integrated unit, if realized in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in the various embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0088] The basic principles, main features and advantages of the present application are shown and described above. Those skilled in the art should understand that the present application is not limited by the above embodiments, and the above embodiments and descriptions in the specification are only preferred examples of the present application and do not limit the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent regulation of power consumption in a smart power plant, characterized in that, The method comprises the following steps: Collecting power plant equipment energy consumption data, environmental parameters, load demand data, equipment operating state data and historical power consumption rate data, aligning multi-source data by combining the DTW algorithm to obtain time series data; Using a GNN network to topologically model the power plant equipment, dividing fixed load equipment and adjustable load equipment, and introducing an attention mechanism to quantify the sensitivity of the power plant equipment to grid fluctuations; Inputting the time series data into a Transformer model, combining the results output by the GNN network for feature extraction, and obtaining a 24-hour load prediction result; Based on the load prediction result, a control scheme is generated by the joint application of a particle swarm optimization algorithm and a DQN network; Converting the control scheme into control instructions, issuing the control instructions, and collecting data after the control for real-time feedback; The GNN network is used to topologically model the power plant equipment, divide fixed load equipment and adjustable load equipment, which comprises: Taking the power plant equipment as nodes and the energy transmission relationship between the equipment as edges, an initial topology graph is constructed; Based on the initial topology graph, the GNN network updates the features of each node by aggregating neighbor node information, models the association relationship of the equipment, and forms a complete power plant equipment topology structure graph; Traverse the nodes in the power plant equipment topology structure graph, analyze the load characteristics of each device, divide the fixed load equipment and the adjustable load equipment, and obtain the classification result; Among them, the fixed load equipment is the equipment with stable operating state and unable to change energy consumption through control, and the adjustable load equipment is the equipment with energy consumption changing with operating parameters; Based on the load prediction result, a control scheme is generated by the joint application of a particle swarm optimization algorithm and a DQN network, which comprises: Taking the 24-hour load prediction result, the adjustable load equipment and the equipment sensitivity as input, initializing the particle swarm, each particle representing a set of control parameter combinations, and calculating the fitness combining the equipment sensitivity; Particles update their positions and velocities by tracking individual optimal solutions and global optimal solutions, moving towards better control parameter combinations, and after 50 iterations, a set of high-fitness control parameter candidates is obtained; Determine the state space and action space of the DQN network, score each control parameter combination in the control parameter candidate set through the experience replay mechanism; According to the scoring results, the control parameter candidate set is screened, and the control parameter combination with the highest score is selected as the final control scheme.

2. The power rate intelligent regulation method for a smart power plant of claim 1, wherein The collection of power plant equipment energy consumption data, environmental parameters, load demand data, equipment operating state data and historical power consumption rate data, the alignment of multi-source data by combining the DTW algorithm, and the obtaining of time series data, comprises: Taking the minute-level timestamp of the equipment operating state data as the reference benchmark, converting data with different sampling frequencies into time series form, and each data point corresponds to a unique time marker; Calculate the distance between different data sequences, find the optimal matching path through dynamic programming, adjust the data with time offset, and obtain time series data with uniform time scale.

3. The power rate intelligent regulation method for a smart power plant of claim 1, wherein The introduction of an attention mechanism to quantify the sensitivity of the power plant equipment to grid fluctuations comprises: In the GNN network, the attention weight of each node to its neighbor node is calculated, and through the LeakyReLU activation function and Softmax normalization, the normalized attention weight is obtained. The normalized attention weight is quantified as the device sensitivity, and the device sensitivity matrix is generated according to the attention weight and the classification result, wherein the row of the matrix represents the device, and the column represents the sensitivity index.

4. The power rate intelligent regulation method for a smart power plant of claim 1, wherein, The time series data is input into the Transformer model, and the results output by the GNN network are combined for feature extraction to obtain the 24-hour load prediction result, including: The time series data is input into the encoder layer of the Transformer model, and the device topology correlation information output by the GNN network is input into the self-attention mechanism module of the encoder as auxiliary features; The encoder first extracts time features based on time series data, then distinguishes and extracts the stable features of fixed load devices and the fluctuation features of adjustable load devices, and generates a comprehensive feature vector that integrates time correlation, device topology correlation and load attributes through multiple layers of processing; The comprehensive feature vector is input into the decoder layer of the Transformer model, and the decoder generates 24-hour load prediction values layer by layer through cross-attention mechanism to finally output the complete 24-hour load prediction result.

5. The intelligent power plant power rate intelligent regulation method of claim 1, wherein, The load prediction result, the combination of the control parameter candidate set and the real-time running state of the device are taken as the state space; The combination of the control parameter candidate set is taken as the selectable action of the action space, and the device correlation label is added to each action combined with the device topology correlation information.

6. A power rate intelligent regulation system for a smart power plant, characterized in that, The system comprises: The acquisition module is used to acquire power plant device energy consumption data, environmental parameters, load demand data, device running state data and historical power consumption rate data, align multiple source data by combining DTW algorithm to obtain time series data; The topology modeling module is used to model the topology of the power plant devices by using the GNN network, to divide the fixed load devices and the adjustable load devices, and to introduce the attention mechanism to quantify the sensitivity of the power plant devices to the grid fluctuations: the power plant devices are taken as nodes, and the energy transmission relationship between the devices is taken as edges to construct an initial topology graph; based on the initial topology graph, the GNN network updates the features of each node by aggregating the neighbor node information to complete the modeling of the device correlation and form a complete power plant device topology structure graph; the nodes in the power plant device topology structure graph are traversed, the load characteristics of each device are analyzed, the fixed load devices and the adjustable load devices are divided, and the classification result is obtained; wherein, the fixed load device is a device with stable running state and unable to change energy consumption through regulation and control, and the adjustable load device is a device with energy consumption changing with running parameters; The feature extraction module is used to input the time series data into the Transformer model, and the results output by the GNN network are combined for feature extraction to obtain the 24-hour load prediction result; The generating module is configured to generate a regulation scheme based on the load prediction result by joint application of a particle swarm optimization algorithm and a DQN network: taking the future 24-hour load prediction result, the adjustable load device, and the device sensitivity as input, initializing the particle swarm, each particle representing a set of regulation parameter combinations, and calculating the fitness in combination with the device sensitivity; the particles update the position and speed by tracking the individual optimal solution and the global optimal solution, and move to better regulation parameter combinations, and after 50 iterations, a set of regulation parameter candidate set with high fitness is obtained; the state space and the action space of the DQN network are determined, and each regulation parameter combination in the regulation parameter candidate set is scored through the experience replay mechanism; the regulation parameter candidate set is screened according to the score result, and the regulation parameter combination with the highest score is selected as the final regulation scheme; The regulation module is configured to convert the regulation scheme into a regulation instruction, issue the regulation instruction, and collect the data after regulation for real-time feedback.

7. A smart power plant power rate intelligent regulation device, characterized in that, The smart power plant power rate intelligent regulation device includes a memory and at least one processor, and the memory stores instructions; the at least one processor calls the instructions in the memory, so that the smart power plant power rate intelligent regulation device executes each step of the smart power plant power rate intelligent regulation method according to any one of claims 1-5.

8. A computer-readable storage medium having stored thereon instructions, the computer-readable storage medium comprising: The instructions are executed by the processor to implement each step of the smart power plant power rate intelligent regulation method according to any one of claims 1-5.

Images