Power supply and distribution method and system based on multi-modal data fusion and adaptive control

Abstract

The invention discloses a power supply and distribution method and system based on multi-modal data fusion and adaptive control, and the method comprises the steps: collecting real-time power supply and distribution data in a power supply and distribution network node, and obtaining power supply and distribution environment data; fusing the real-time power supply and distribution data and the power supply and distribution environment data by using a quantum heuristic multi-modal fusion algorithm to obtain fused power supply and distribution data; establishing an LSTM-CNN machine learning model based on the LSTM long and short term memory network and the CNN; inputting the fused power supply and distribution data into an LSTM-CNN machine learning model for training to obtain electric load prediction data; and based on the adaptive control strategy, sending a regulation and control instruction to the electric load equipment according to the electric load prediction data. And voltage fluctuation, flicker and harmonic distortion are effectively suppressed through real-time dynamic regulation and control, and voltage deviation and power factors of a user terminal are guaranteed to be optimal.

Description

Technical Field

[0001] The present invention relates to the field of power supply and distribution, and in particular to a power supply and distribution method and system based on multi-modal data fusion and adaptive control. Background Art

[0002] The traditional power supply and distribution model relies on isolated electrical monitoring data, and has a weak control over the real-time status of the power grid. In the face of sudden changes in power load, such as the surge in regional power consumption caused by e-commerce shopping festivals and the blowout of air conditioning load caused by high temperatures in summer, the response is often delayed, resulting in voltage drops and line overloads. Distributed energy is booming, and a large number of photovoltaic and wind power are connected in a scattered manner. Traditional control methods are stretched to the limit and it is difficult to tame the intermittent and volatile output, resulting in disordered power grid flow and deterioration of power quality. Although the existing power grid has a massive data "rich mine" covering multimodal information such as electrical parameters, equipment images, and user energy consumption curves, it lacks a deep fusion mechanism and is like a pile of loose sand. It has failed to transform potential value into decision-making advantages and cannot meet the current power grid's stringent demands for refined and agile management and control. Summary of the invention

[0003] The purpose of the present invention is to solve the above problems and to design a power supply and distribution method and system based on multimodal data fusion and adaptive control.

[0004] To achieve the above object, the technical solution of the present invention is that, further, in the above-mentioned power supply and distribution method based on multimodal data fusion and adaptive control, the adaptive control power supply and distribution method comprises the following steps:

[0005] Collect real-time power supply and distribution data in power supply and distribution network nodes, and obtain power supply and distribution environment data;

[0006] The real-time power supply and distribution data and the power supply and distribution environment data are fused using a quantum-inspired multimodal fusion algorithm to obtain fused power supply and distribution data;

[0007] Establish LSTM-CNN machine learning model based on LSTM long short-term memory network and CNN convolutional neural network;

[0008] Inputting the fused power supply and distribution data into the LSTM-CNN machine learning model for training to obtain power load forecast data;

[0009] Based on the adaptive control strategy, control instructions are sent to the electric load equipment according to the electric load prediction data.

[0010] Furthermore, in the above-mentioned power supply and distribution method based on multimodal data fusion and adaptive control, the collecting of real-time power supply and distribution data in the power supply and distribution network nodes and obtaining of power supply and distribution environment data include:

[0011] The power supply and distribution network nodes at least include the substation outlet, the distribution room bus, and the user's incoming line, where high-precision sensor clusters are deployed to collect real-time power supply and distribution data in the power supply and distribution network nodes;

[0012] The real-time power supply and distribution data at least includes voltage transformer data, current transformer data, power factor meter, voltage parameters, current parameters, active power parameters, reactive power parameters, and power factor electrical parameters;

[0013] Power supply and distribution environment data is obtained, where the power supply and distribution environment data at least includes ambient temperature data, humidity data, and wind speed data.

[0014] Furthermore, in the above-mentioned power supply and distribution method based on multimodal data fusion and adaptive control, the real-time power supply and distribution data and the power supply and distribution environment data are fused by using a quantum-inspired multimodal fusion algorithm to obtain fused power supply and distribution data, including:

[0015] Based on the quantum-inspired multimodal fusion algorithm, quantum bit encoding is used to map real-time power supply and distribution data and power supply and distribution environment data into quantum state space;

[0016] The characteristics of quantum entanglement are used to mine implicit cross-modal correlations, capture power fluctuations, equipment micro-damages and user habits, and obtain integrated power supply and distribution data.

[0017] Furthermore, in the above-mentioned power supply and distribution method based on multimodal data fusion and adaptive control, the LSTM-CNN machine learning model is established based on the LSTM long short-term memory network and the CNN convolutional neural network, including:

[0018] Establish LSTM-CNN machine learning model based on LSTM long short-term memory network and CNN convolutional neural network;

[0019] Input the integrated power supply and distribution data into the LSTM-CNN machine learning model for training to predict the power load trend within the next 1 minute to 2 hours;

[0020] The fault tree analysis method and state estimation technology are used to analyze the current operating conditions of the power grid and obtain the power load prediction data.

[0021] Furthermore, in the above-mentioned power supply and distribution method based on multimodal data fusion and adaptive control, the adaptive control strategy is based on which a control instruction is sent to the electric load device according to the electric load forecast data, including:

[0022] Based on the forecast data of power load, if the power load is predicted to rise sharply, the tap position of the substation transformer is adjusted to raise the output voltage; the switching capacity of the reactive power compensation device is dynamically adjusted;

[0023] For industrial flexible production workshops, control instructions are sent to interruptible load equipment to reduce electricity consumption in non-critical processes;

[0024] If a short circuit occurs suddenly in the power grid, the faulty line will be cut off, and the surrounding distributed power sources and energy storage systems will be linked to the grid for emergency power supply. The characteristics of distributed power sources and energy storage functions will be used to fill the power gap and stabilize the frequency and voltage of the power grid.

[0025] Further, in the power supply and distribution system based on multimodal data fusion and adaptive control, the adaptively controlled power supply and distribution system includes the following modules:

[0026] A data collection module is used to collect real-time power supply and distribution data in the power supply and distribution network nodes and obtain power supply and distribution environment data;

[0027] A data fusion module, used to fuse the real-time power supply and distribution data with the power supply and distribution environment data using a quantum-inspired multimodal fusion algorithm to obtain fused power supply and distribution data;

[0028] Model building module, used to build LSTM-CNN machine learning model based on LSTM long short-term memory network and CNN convolutional neural network;

[0029] An electric load prediction module, used for inputting the fused power supply and distribution data into the LSTM-CNN machine learning model for training to obtain electric load prediction data;

[0030] The strategy generation module is used to send control instructions to the electric load equipment based on the adaptive control strategy and according to the electric load prediction data.

[0031] Furthermore, in the power supply and distribution system based on multimodal data fusion and adaptive control, the data fusion module includes the following submodules:

[0032] A mapping submodule is used to map real-time power supply and distribution data and power supply and distribution environment data into quantum state space using quantum bit encoding based on a quantum-inspired multimodal fusion algorithm;

[0033] The correlation submodule is used to utilize the characteristics of quantum entanglement to mine cross-modal implicit correlations, capture power fluctuations, equipment micro-damages and user habits, and obtain integrated power supply and distribution data.

[0034] Furthermore, in the power supply and distribution system based on multimodal data fusion and adaptive control, the electric load prediction module includes the following submodules:

[0035] Establish a submodule to build an LSTM-CNN machine learning model based on the LSTM long short-term memory network and the CNN convolutional neural network;

[0036] A prediction submodule, used for inputting the integrated power supply and distribution data into the LSTM-CNN machine learning model for training, and predicting the power load trend within the next 1 minute to 2 hours;

[0037] The obtained submodule is used to analyze the current operating conditions of the power grid using the fault tree analysis method and state estimation technology to obtain the electric load prediction data.

[0038] Furthermore, in the power supply and distribution system based on multimodal data fusion and adaptive control, the strategy generation module includes the following submodules:

[0039] The judgment submodule is used to make judgments based on the power load forecast data. If the power load is predicted to rise sharply, the tap position of the substation transformer is adjusted to increase the output voltage; the switching capacity of the reactive compensation device is dynamically adjusted;

[0040] The instruction generation submodule is used to send control instructions to interruptible load equipment in industrial flexible production workshops to reduce electricity consumption in non-critical processes;

[0041] The linkage submodule is used to cut off the faulty line if a short circuit occurs in the power grid, and to link the surrounding distributed power sources and energy storage systems to emergency grid-connected power supply. It uses the characteristics of distributed power sources and energy storage functions to fill the power gap and stabilize the frequency and voltage of the power grid.

[0042] Its beneficial effects are that by collecting real-time power supply and distribution data in the power supply and distribution network nodes and obtaining power supply and distribution environment data; using the quantum-inspired multimodal fusion algorithm to fuse the real-time power supply and distribution data and the power supply and distribution environment data to obtain fused power supply and distribution data; establishing an LSTM-CNN machine learning model based on the LSTM long short-term memory network and the CNN convolutional neural network; inputting the fused power supply and distribution data into the LSTM-CNN machine learning model for training to obtain power load forecast data; based on the adaptive control strategy, sending control instructions to the power load equipment according to the power load forecast data. 1. Improve the quality of electric energy: Real-time dynamic control effectively suppresses voltage fluctuations, flickers and harmonic distortion, ensures that the voltage deviation at the user end is within ±5%, and the power factor is excellent, creating a high-quality power environment for the stable operation of precision electronic equipment and medical instruments. 2. Enhance the reliability of the power grid: It can respond quickly to complex working conditions and sudden failures, and has outstanding self-healing recovery capabilities, reducing the incidence of power outages and reducing economic losses caused by power outages. 3. Optimize energy utilization: Accurately adapt to load demand, reduce peak-to-valley differences in the power grid, fully tap the potential of distributed energy, and help achieve energy conservation and emission reduction in the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the following detailed description of the preferred embodiment.The drawings are only for the purpose of illustrating the preferred embodiments and are not to be construed as limiting the invention.

[0044] Figure 1 It is a schematic diagram of a first embodiment of a power supply and distribution method based on multimodal data fusion and adaptive control in an embodiment of the present invention;

[0045] Figure 2 Schematic diagram of a second embodiment of a power supply and distribution method based on multimodal data fusion and adaptive control in an embodiment of the present invention;

[0046] Figure 3 It is a schematic diagram of a first embodiment of a power supply and distribution system based on multimodal data fusion and adaptive control in an embodiment of the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0048] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0049] The present invention will be described in detail below in conjunction with the accompanying drawings. Figure 1 As shown, a power supply and distribution method based on multimodal data fusion and adaptive control, the adaptive control power supply and distribution method comprises the following steps:

[0050] Step 101: Collect real-time power supply and distribution data in the power supply and distribution network nodes, and obtain power supply and distribution environment data;

[0051] Specifically, in this embodiment, the power supply and distribution network nodes at least include high-precision sensor clusters deployed at the substation outlet, the distribution room bus, and the user entrance to collect real-time power supply and distribution data in the power supply and distribution network nodes; the real-time power supply and distribution data at least include voltage transformer data, current transformer data, power factor meter, voltage parameters, current parameters, active power parameters, reactive power parameters, and power factor electrical parameters; the power supply and distribution environment data is obtained, and the power supply and distribution environment data at least includes ambient temperature data, humidity data, and wind speed data.

[0052] Step 102: using a quantum-inspired multimodal fusion algorithm to fuse the real-time power supply and distribution data and the power supply and distribution environment data to obtain fused power supply and distribution data;

[0053] Specifically, in this embodiment, real-time power supply and distribution data cleaning: the real-time power supply and distribution data collected from multiple devices such as substations, distribution rooms, distribution cabinets at all levels, and smart meters will inevitably be mixed with various types of noise and abnormal values. For example, due to electromagnetic interference, voltage and current data may appear spikes and glitches instantly; or when communication fails, some time period data is missing. To this end, a denoising algorithm based on wavelet transform is used to accurately identify and remove high-frequency noise components; for the problem of missing data, a combination of linear interpolation and multiple filling is used to reasonably fill in the missing values ​​according to the law of surrounding time period data and the correlation of data of the same type of equipment, to ensure data continuity and accuracy, and lay a solid foundation for subsequent fusion. Standardization of power supply and distribution environment data: The power supply and distribution environment data comes from a wide range of sources, including meteorological environment information monitored by temperature and humidity sensors, terrain and soil resistivity data around the line provided by the geographic information system (GIS), and quantitative data converted by image recognition of line appearance and surrounding vegetation coverage captured by high-definition cameras carried by line inspection drones. In view of the huge differences in the dimensions and value ranges of data from different sources, the Z-score standardization method is used to uniformly transform various environmental data into a standard normal distribution interval with a mean of 0 and a standard deviation of 1, eliminating the fusion bias caused by inconsistent data scales; at the same time, principal component analysis (PCA) is used to reduce the dimension of the features extracted from image data, retain key environmental feature information, and improve the efficiency of subsequent algorithm processing. Quantum-inspired model construction and initialization quantum bit encoding: Drawing on the unique encoding method of quantum bits in quantum computing, the pre-processed real-time power supply and distribution data and power supply and distribution environment data are mapped to quantum bit sequences respectively. Taking real-time electrical parameters such as voltage and current as an example, after their numerical values ​​are discretized, they correspond to different superposition states of quantum bits, giving each data dimension a quantum state representation form; the same is true for environmental data, such as temperature values, which are encoded into the quantum bit system. With the help of the superposition characteristics of quantum bits that can simultaneously represent multiple states, complex data information can be cleverly accommodated, and potential data associations can be mined. Compared with traditional binary encoding, the data expression capability is greatly expanded. Model parameter setting: Carefully adjust the parameters of the quantum-inspired multimodal fusion algorithm according to the scale, complexity and data real-time requirements of the power supply and distribution system. Determine the probability of quantum gate operation, which is directly related to the evolution rate and diversity of the quantum bit state, affecting the efficiency of the algorithm optimization; reasonably set the population size, to ensure that there are enough potential solutions to traverse the optimal data fusion solution, and avoid excessive consumption of computing resources and slow operations due to a large population; at the same time, clarify the iteration termination conditions, comprehensively consider the improvement in data fusion accuracy, and the lack of significant improvement in several consecutive rounds of iterations, so as to prevent the algorithm from falling into an invalid cycle, accurately control the operation time, and match the real-time control rhythm of power supply and distribution. Multimodal fusion iterative operation quantum state evolution and data interaction: After starting the algorithm, the quantum bit continues to evolve according to the set quantum gate operation rules to simulate the dynamic process of the quantum system.During this period, the quantum bits representing real-time power supply and distribution data and power supply and distribution environment data interact frequently, and with the help of quantum entanglement characteristics, the data modal barriers are broken to achieve deep information sharing. For example, the real-time current overload information of a certain outgoing line of the substation is instantly linked to the high temperature and humid environment data of the surrounding lines through quantum entanglement, providing a comprehensive perspective for judging the potential fault risks of the line; each round of iteration measures and evaluates the state of the quantum bits and extracts the intermediate data results after fusion. Fitness function evaluation and screening: Construct the fitness function of the fusion effect, and comprehensively consider the contribution of the fused data to key indicators such as the accuracy of power supply and distribution system stability prediction, the timeliness of fault warning, and the accuracy of energy loss assessment. After each round of quantum state evolution and data interaction is completed, the current fusion result is substituted into the fitness function for evaluation; based on the level of fitness, the elite retention strategy is used to screen high-quality quantum bit combinations and eliminate inferior solutions; the selected excellent solutions serve as the "seeds" for the next round of iterations, continuously guiding the algorithm to approach a better fusion solution, and after multiple rounds of iterations, it gradually converges to the global optimum, obtaining integrated power supply and distribution data that accurately reflects the overall operation of the power supply and distribution system, accurately empowering subsequent power supply and distribution scheduling and operation and maintenance decisions.

[0054] The introduction of quantum-inspired multimodal fusion algorithms is very different from conventional linear fusion ideas. With the help of quantum bit encoding, continuous electrical data such as voltage and current, discrete pixel information of equipment inspection images, and user electricity time series data are mapped to quantum state space, and the characteristics of quantum entanglement are used to mine cross-modal implicit associations, accurately capture the "hidden link" between instantaneous power fluctuations, equipment micro-damages and user habit mutations, and achieve unprecedented deep fusion accuracy. The data fusion dimension is expanded by more than 30%, and the accuracy rate of early identification of fault hazards has soared to more than 95%.

[0055] Step 103: Establish an LSTM-CNN machine learning model based on the LSTM long short-term memory network and the CNN convolutional neural network;

[0056] Specifically, in this embodiment, an LSTM-CNN machine learning model is established based on the LSTM long short-term memory network and the CNN convolutional neural network; the fused power supply and distribution data is input into the LSTM-CNN machine learning model for training to predict the power load trend in the next 1 minute to 2 hours; the fault tree analysis method and state estimation technology are used to analyze the current operating conditions of the power grid to obtain power load prediction data.

[0057] Step 104: input the integrated power supply and distribution data into the LSTM-CNN machine learning model for training to obtain power load prediction data;

[0058] Specifically, in this embodiment, a judgment is made based on the electric load forecast data. If the predicted electric load is going to rise sharply, the tap position of the substation transformer is adjusted to raise the output voltage; the switching capacity of the reactive compensation device is dynamically adjusted; for industrial flexible production workshops, control instructions are sent to interruptible load equipment to reduce electricity consumption in non-critical processes; if a short circuit occurs in the power grid, the faulty line is cut off, and the surrounding distributed power sources and energy storage systems are linked to emergency grid-connected power supply, using the characteristics of distributed power sources and energy storage functions to fill the power gap and stabilize the frequency and voltage of the power grid.

[0059] Step 105: Based on the adaptive control strategy, control instructions are sent to the electric load equipment according to the electric load forecast data.

[0060] Specifically, in this embodiment, the data acquisition module is used to collect real-time power supply and distribution data in the power supply and distribution network nodes, and obtain power supply and distribution environment data; the data fusion module is used to use the quantum-inspired multimodal fusion algorithm to fuse the real-time power supply and distribution data and the power supply and distribution environment data to obtain fused power supply and distribution data; the model building module is used to establish an LSTM-CNN machine learning model based on the LSTM long short-term memory network and the CNN convolutional neural network; the electric load prediction module is used to input the fused power supply and distribution data into the LSTM-CNN machine learning model for training to obtain electric load prediction data; the strategy generation module is used to send control instructions to the electric load equipment according to the electric load prediction data based on the adaptive control strategy.

[0061] Its beneficial effects are that by collecting real-time power supply and distribution data in the power supply and distribution network nodes and obtaining power supply and distribution environment data; using the quantum-inspired multimodal fusion algorithm to fuse the real-time power supply and distribution data and the power supply and distribution environment data to obtain fused power supply and distribution data; establishing an LSTM-CNN machine learning model based on the LSTM long short-term memory network and the CNN convolutional neural network; inputting the fused power supply and distribution data into the LSTM-CNN machine learning model for training to obtain the power load forecast data; based on the adaptive control strategy, sending control instructions to the power load equipment according to the power load forecast data. 1. Improve the quality of electric energy: Real-time dynamic control effectively suppresses voltage fluctuations, flickers and harmonic distortion, ensures that the voltage deviation at the user end is within ±5%, and the power factor is excellent, creating a high-quality power environment for the stable operation of precision electronic equipment and medical instruments. 2. Enhance the reliability of the power grid: It can respond quickly to complex working conditions and sudden failures, and has outstanding self-healing recovery capabilities, reducing the incidence of power outages and reducing economic losses caused by power outages. 3. Optimize energy utilization: Accurately adapt to load demand, reduce peak-to-valley differences in the power grid, fully tap the potential of distributed energy, and help achieve energy conservation and emission reduction in the power system.

[0062] See also Figure 2In the power supply and distribution method based on multimodal data fusion and adaptive control, establishing an LSTM-CNN machine learning model based on LSTM long short-term memory network and CNN convolutional neural network includes the following steps:

[0063] Step 201: Establish an LSTM-CNN machine learning model based on the LSTM long short-term memory network and the CNN convolutional neural network;

[0064] Step 202: Input the integrated power supply and distribution data into the LSTM-CNN machine learning model for training to predict the power load trend within the next 1 minute to 2 hours;

[0065] Step 203: Analyze the current operating conditions of the power grid using the fault tree analysis method and state estimation technology to obtain power load prediction data.

[0066] The above is an introduction to the embodiment of the power supply and distribution method based on multimodal data fusion and adaptive control of the present invention. Figure 3 In the power supply and distribution system based on multimodal data fusion and adaptive control, the adaptive control power supply and distribution system includes the following modules:

[0067] A data collection module is used to collect real-time power supply and distribution data in the power supply and distribution network nodes and obtain power supply and distribution environment data;

[0068] A data fusion module is used to fuse real-time power supply and distribution data with power supply and distribution environment data using a quantum-inspired multimodal fusion algorithm to obtain fused power supply and distribution data;

[0069] Model building module, used to build LSTM-CNN machine learning model based on LSTM long short-term memory network and CNN convolutional neural network;

[0070] The electric load prediction module is used to input the integrated power supply and distribution data into the LSTM-CNN machine learning model for training to obtain the electric load prediction data;

[0071] The strategy generation module is used to send control instructions to the electric load equipment based on the adaptive control strategy and the electric load forecast data.

[0072] Specifically, this embodiment also includes:

[0073] Load forecasting and operating condition assessment:

[0074] Principles of machine learning model construction and integration: CCU (central control unit) is the "intelligent brain" of the entire power supply and distribution system. The advanced machine learning model carefully constructed inside is not a simple model superposition, but a deep integration of the unique advantages of long short-term memory network (LSTM) and convolutional neural network (CNN). LSTM is good at capturing long-term and short-term dependencies in time series data. Given that the power load fluctuates over time and presents complex periodicity and trends, the hourly, daily, and monthly load curves in the past few years contain rich rules. LSTM can accurately track the daily peak and valley power consumption habits, seasonal power consumption fluctuations, and load mutation traces caused by special events; for example, it can keenly detect the high load at night caused by the continuous cooling of air conditioners during the high temperature period in summer, or the low power consumption caused by the shutdown of factories and residents returning home during the Spring Festival holiday. CNN, relying on its powerful convolution kernel, is adept at two-dimensional data processing, converting real-time collected electrical parameters (real-time waveform images of voltage and current) and meteorological information (grid temperature, humidity, and air pressure distribution maps) into a two-dimensional matrix suitable for processing, and efficiently extracting local features, such as accurately identifying the subtle distortion characteristics of current caused by the dampness of the line and the instantaneous increase in resistance due to heavy rain, or the abnormal pattern of voltage fluctuation caused by branches touching the line due to sudden strong winds in a certain area. The two are integrated through a special neural network structure, and data flows interactively between different levels. The time trend information sorted out by the early LSTM is input into CNN to further enhance the accuracy of feature extraction. The key features output by CNN reversely assist LSTM to calibrate the time-dependent weights to achieve the effect of 1+1>2.

[0075] In the model training phase, massive amounts of historical electricity consumption data were divided into training sets, validation sets, and test sets. A small-batch gradient descent algorithm was used, combined with an adaptive learning rate adjustment strategy, and hundreds of rounds of iterative training were performed. At the same time, the early stopping method was introduced to prevent the model from overfitting, ensuring that the model can accurately fit the laws of historical data and have excellent generalization capabilities, and accurately predict the ever-changing electricity load trends in the future.

[0076] Multivariate data fusion details:

[0077] The electrical parameters collected in real time are continuously fed into the CCU multiple times per second via high-speed communication links. These data are instantly analyzed, verified, and regularized by professional data analysis software to remove abnormal values. Meteorological information comes from data connection with professional meteorological agencies and monitoring by local micro-meteorological stations, covering meteorological micro-environment data within a radius of several kilometers. Before data fusion, normalization and standardization techniques are used to unify the scale of data of different magnitudes and dimensions. Based on feature engineering, exclusive feature vectors are created for different types of data, such as designing features for electrical parameters that reflect the power fluctuation rate and three-phase imbalance, and constructing features related to sudden temperature changes and humidity saturation for meteorological data, enriching the amount of information contained in the data in all directions. Finally, various features are seamlessly spliced ​​through the fully connected neural network layer, and input into the core area of ​​the fusion model to start the precise prediction process.

[0078] Working condition assessment technology practice:

[0079] When constructing the fault tree analysis method, serious faults such as power outages and equipment damage are taken as top events. According to detailed fault mechanisms such as line aging, short circuits, overloads, equipment overheating, and insulation breakdown, intermediate events and bottom events are disassembled layer by layer, and each event is given an accurate probability of occurrence (derived from historical fault statistics and equipment reliability manuals), and a visual fault tree diagram is drawn; when real-time electrical parameters and environmental monitoring data flow in, Bayesian reasoning is used to reversely trace back along the fault tree to accurately locate potential fault points and weak links under current working conditions. State estimation technology uses the weighted least squares method, based on limited real-time data obtained by the measurement devices of substations and distribution rooms, combined with the grid topology and line parameters to establish state equations, and solve state variables such as voltage amplitude and phase angle of all nodes in the grid; compared with the normal operating threshold, the current grid deviates from the healthy state to accurately quantify the degree, comprehensively evaluate the probability of fault occurrence and the geographical area and power equipment range that may be affected once a fault occurs, and generate a detailed working condition assessment report to lay a solid foundation for subsequent control strategies.

[0080] Adaptive control strategy generation

[0081] Analysis of strategies to cope with load increase:

[0082] When the load forecasting model issues a warning of a sharp increase in power load, the signal instantly activates a series of interlocking control actions. Adjusting the position of the transformer tap in the substation is the first measure. The CCU remotely controls the substation automation equipment to drive the electric mechanism of the tap changer to move accurately; with the help of motor drive and gear transmission, the tap position is switched smoothly to increase the output voltage; this process is monitored and fed back in real time by high-precision voltage sensors, combined with the intelligent PID (proportional-integral-differential) control algorithm to fine-tune the action amplitude to ensure that the voltage is raised just right and avoid overvoltage hazards.

[0083] The deployment of reactive power compensation devices is also closely linked. SVC (static reactive power compensator) relies on thyristors to quickly switch capacitors and reactors, while SVG (static reactive power generator) uses switchable power electronic devices to flexibly emit or absorb reactive power. CCU accurately calculates the required reactive power compensation based on real-time power factor monitoring data and issues millisecond-level switching instructions. After the device responds, it verifies the compensation effect through real-time data feedback from reactive power flow monitoring points, and dynamically fine-tunes the switching strategy to ensure that the power factor is stably maintained above 0.95, reduce reactive power losses in the power grid, and improve power transmission efficiency.

[0084] For industrial flexible production workshops, CCU is deeply interconnected with the workshop manufacturing execution system (MES) to grasp the real-time power consumption and equipment operation progress of each production process; once the load is tight, non-critical processes, such as product packaging and material temporary storage area equipment, are quickly screened out, and control instructions are sent to their programmable logic controllers (PLCs); PLCs suspend and delay the operation of some equipment in an orderly manner according to the instructions to smooth the peak power load of the workshop; during this period, the workshop power monitoring system provides real-time feedback on the power consumption reduction of equipment, and CCU optimizes the control instructions as needed to achieve precise load control.

[0085] Panorama of emergency response to short circuit faults:

[0086] At the moment of a sudden short circuit in the power grid, the protection device at the head end of the faulty line (such as a relay protection relay) will immediately operate, using the current mutation and differential protection principles to accurately identify the fault within milliseconds, triggering a trip command to cut off the faulty line and isolate the source of the fault; at the same time, the CCU will simultaneously initiate an emergency response mechanism, relying on the pre-built distributed power supply and energy storage system management and control network to issue emergency grid-connected power supply commands to surrounding distributed photovoltaic power stations, small wind farms, and energy storage battery packs.

[0087] Distributed power sources rely on their natural advantage of "generating and using immediately". Photovoltaic panels can output direct current immediately without preheating under light conditions, and are efficiently converted into alternating current for grid connection through inverters. Wind turbine blades rotate faster against the wind, quickly increasing power generation. The energy storage system gives full play to its expertise in "peak shaving and valley filling". Lithium battery packs and supercapacitors instantly release stored energy to make up for the power supply lost due to line tripping. CCU monitors grid frequency and voltage fluctuations in real time, and regulates the output of distributed power sources and the charging and discharging rate of energy storage with the help of smart grid control technology. It coordinates the power distribution between various power generation and energy storage units through the distributed energy management system to ensure that the grid frequency is stable at 50Hz±0.2Hz and the voltage deviation is controlled within ±5%, so as to survive the critical period of fault repair and quickly restore the normal power supply order of the grid.

[0088] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A power supply and distribution method based on multimodal data fusion and adaptive control, characterized in that: The adaptively controlled power supply and distribution method comprises the following steps: Collect real-time power supply and distribution data in power supply and distribution network nodes, and obtain power supply and distribution environment data; The real-time power supply and distribution data and the power supply and distribution environment data are fused by using a quantum-inspired multimodal fusion algorithm to obtain fused power supply and distribution data; Establish LSTM-CNN machine learning model based on LSTM long short-term memory network and CNN convolutional neural network; Inputting the fused power supply and distribution data into the LSTM-CNN machine learning model for training to obtain power load forecast data; Based on the adaptive control strategy, control instructions are sent to the electric load equipment according to the electric load prediction data.

2. The power supply and distribution method based on multimodal data fusion and adaptive control according to claim 1, characterized in that: The collecting of real-time power supply and distribution data in the power supply and distribution network nodes and obtaining power supply and distribution environment data includes: The power supply and distribution network nodes at least include the substation outlet, the distribution room bus, and the user's incoming line, where high-precision sensor clusters are deployed to collect real-time power supply and distribution data in the power supply and distribution network nodes; The real-time power supply and distribution data at least includes voltage transformer data, current transformer data, power factor meter, voltage parameters, current parameters, active power parameters, reactive power parameters, and power factor electrical parameters; Power supply and distribution environment data is obtained, where the power supply and distribution environment data at least includes ambient temperature data, humidity data, and wind speed data.

3. The power supply and distribution method based on multimodal data fusion and adaptive control according to claim 1, characterized in that: The real-time power supply and distribution data and the power supply and distribution environment data are fused by using a quantum-inspired multimodal fusion algorithm to obtain fused power supply and distribution data, including: Based on the quantum-inspired multimodal fusion algorithm, quantum bit encoding is used to map real-time power supply and distribution data and power supply and distribution environment data into quantum state space; The characteristics of quantum entanglement are used to mine implicit cross-modal correlations, capture power fluctuations, equipment micro-damages and user habits, and obtain integrated power supply and distribution data.

4. The power supply and distribution method based on multimodal data fusion and adaptive control according to claim 1, characterized in that: The LSTM-CNN machine learning model is established based on the LSTM long short-term memory network and the CNN convolutional neural network, including: Establish LSTM-CNN machine learning model based on LSTM long short-term memory network and CNN convolutional neural network; Input the integrated power supply and distribution data into the LSTM-CNN machine learning model for training to predict the power load trend within the next 1 minute to 2 hours; The fault tree analysis method and state estimation technology are used to analyze the current operating conditions of the power grid and obtain the power load prediction data.

5. The power supply and distribution method based on multimodal data fusion and adaptive control according to claim 1, characterized in that: The method of sending a control instruction to the electric load device based on the adaptive control strategy according to the electric load prediction data includes: Based on the forecast data of power load, if the power load is predicted to rise sharply, the tap position of the substation transformer is adjusted to raise the output voltage; the switching capacity of the reactive power compensation device is dynamically adjusted; For industrial flexible production workshops, control instructions are sent to interruptible load equipment to reduce electricity consumption in non-critical processes; If a short circuit occurs suddenly in the power grid, the faulty line will be cut off, and the surrounding distributed power sources and energy storage systems will be linked to the grid for emergency power supply. The characteristics of distributed power sources and energy storage functions will be used to fill the power gap and stabilize the frequency and voltage of the power grid.

6. A power supply and distribution system based on multimodal data fusion and adaptive control, characterized in that: The system includes the following modules: A data collection module is used to collect real-time power supply and distribution data in the power supply and distribution network nodes and obtain power supply and distribution environment data; A data fusion module, used to fuse the real-time power supply and distribution data with the power supply and distribution environment data using a quantum-inspired multimodal fusion algorithm to obtain fused power supply and distribution data; Model building module, used to build LSTM-CNN machine learning model based on LSTM long short-term memory network and CNN convolutional neural network; An electric load prediction module, used for inputting the fused power supply and distribution data into the LSTM-CNN machine learning model for training to obtain electric load prediction data; The strategy generation module is used to send control instructions to the electric load equipment based on the adaptive control strategy and according to the electric load prediction data.

7. The power supply and distribution system based on multimodal data fusion and adaptive control according to claim 6, characterized in that: The data fusion module includes the following submodules: A mapping submodule is used to map real-time power supply and distribution data and power supply and distribution environment data into quantum state space using quantum bit encoding based on a quantum-inspired multimodal fusion algorithm; The correlation submodule is used to utilize the characteristics of quantum entanglement to mine cross-modal implicit correlations, capture power fluctuations, equipment micro-damages and user habits, and obtain integrated power supply and distribution data.

8. The power supply and distribution system based on multimodal data fusion and adaptive control according to claim 6, characterized in that: The electric load prediction module includes the following submodules: Establish a submodule to build an LSTM-CNN machine learning model based on the LSTM long short-term memory network and the CNN convolutional neural network; A prediction submodule, used for inputting the integrated power supply and distribution data into the LSTM-CNN machine learning model for training, and predicting the power load trend within the next 1 minute to 2 hours; The obtained submodule is used to analyze the current operating conditions of the power grid using the fault tree analysis method and state estimation technology to obtain the electric load prediction data.

9. The power supply and distribution system based on multimodal data fusion and adaptive control according to claim 6, characterized in that: The strategy generation module includes the following submodules: The judgment submodule is used to make judgments based on the power load forecast data. If the power load is predicted to rise sharply, the tap position of the substation transformer is adjusted to increase the output voltage; the switching capacity of the reactive compensation device is dynamically adjusted; The instruction generation submodule is used to send control instructions to interruptible load equipment in industrial flexible production workshops to reduce electricity consumption in non-critical processes; The linkage submodule is used to cut off the faulty line if a short circuit occurs in the power grid, and to link the surrounding distributed power sources and energy storage systems to emergency grid-connected power supply. It uses the characteristics of distributed power sources and energy storage functions to fill the power gap and stabilize the frequency and voltage of the power grid.

Images