Dispatching algorithm optimization method and system based on power twin

Abstract

The present invention discloses a scheduling algorithm optimization method and system based on power twins, which belongs to the field of digital twins, wherein the method includes: establishing a digital twin model of a power plant, establishing an equipment stability evaluation function; obtaining power demand characteristics; conducting environmental data monitoring of clean power generation, establishing a real-time environmental data set and a predicted environmental data set; initializing channels of intelligent scheduling channels based on power demand characteristics; predicting power generation parameters of clean equipment, generating stable operation data according to the prediction results and the equipment stability evaluation function; generating scheduling decision results, and completing the supply scheduling optimization of the power plant. This application solves the technical problem in the prior art that it is difficult to accurately predict and stably match clean power generation and power grid load based on traditional models, resulting in low power grid scheduling accuracy and poor coordination, and achieves the technical effect of real-time collaborative optimization of clean power generation and power grid load and improving power grid scheduling accuracy and coordination by building digital twins and intelligent scheduling channels.

Description

Technical Field

[0001] The present invention relates to the field of digital twins, and in particular to a scheduling algorithm optimization method and system based on power twins. Background Art

[0002] With the emphasis on clean energy, renewable energy such as solar energy and wind power has been actively developed. However, the existing power grid dispatching uses traditional models to predict the output power of clean power generation equipment. If the historical output power is used for prediction, it may cause large fluctuations in clean power generation, resulting in low power grid dispatching accuracy. In addition, the existing power grid is based on extensive load matching methods and lacks adaptive optimization models, making it difficult to achieve economic operation, resulting in poor coordination between the power grid and clean equipment. Summary of the invention

[0003] This application provides a scheduling algorithm optimization method and system based on power twins, aiming to solve the technical problem in the prior art that it is difficult to accurately predict and stably match clean power generation and grid load based on traditional models, resulting in low grid scheduling accuracy and poor coordination.

[0004] In view of the above problems, the present application provides a scheduling algorithm optimization method and system based on power twins.

[0005] The first aspect disclosed in the present application provides a scheduling algorithm optimization method based on power twins, which includes: establishing a digital twin model of a power plant, the digital twin model is constructed by extracting equipment data of the power plant, and the equipment data includes equipment parameter data and equipment historical working data; performing power source identification, and establishing an equipment stability evaluation function corresponding to the power source through the digital twin model; obtaining power demand characteristics, wherein the power demand characteristics are predictive characteristics; monitoring environmental data of clean power generation through monitoring sensors, and establishing real-time environmental data sets and predictive environmental data sets; inputting the power demand characteristics into an intelligent scheduling channel to initialize the channel of the intelligent scheduling channel; predicting the power generation parameters of clean equipment based on the real-time environmental data and the predictive environmental data sets, and generating operating stability data based on the prediction results and the equipment stability evaluation function; inputting the operating stability data, the real-time environmental data set and the predictive environmental data set into the intelligent scheduling channel to generate a scheduling decision result; and completing the supply scheduling optimization of the power plant based on the scheduling decision result.

[0006] Another aspect disclosed in the present application provides a scheduling algorithm optimization system based on power twins, which includes: a twin model establishment module, which is used to establish a digital twin model of a power plant, and the digital twin model is constructed by extracting equipment data of the power plant, and the equipment data includes equipment parameter data and equipment historical working data; an evaluation function establishment module, which is used to perform power source identification and establish an equipment stability evaluation function corresponding to the power source through the digital twin model; a demand feature acquisition module, which is used to obtain power demand characteristics, wherein the power demand characteristics are prediction characteristics; an environmental data monitoring module, which is used to monitor the environmental data of clean power generation through monitoring sensors, and establish a real-time environmental data set and a predicted environmental data set; a channel initialization module, which is used to input the power demand characteristics into an intelligent scheduling channel to initialize the channel of the intelligent scheduling channel; an operation stability data module, which is used to predict the power generation parameters of clean equipment according to the real-time environmental data and the predicted environmental data set, and generate operation stability data according to the prediction results and the equipment stability evaluation function; a scheduling decision result module, which is used to input the operation stability data, the real-time environmental data set and the predicted environmental data set into the intelligent scheduling channel to generate a scheduling decision result; a supply scheduling optimization module, which is used to complete the supply scheduling optimization of the power plant according to the scheduling decision result.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0008] The digital twin model of the power plant is constructed by using the equipment data of the power plant to provide a basis for establishing evaluation functions, etc.; power source identification is performed to distinguish different sources of power to support power dispatch optimization; an equipment stability evaluation function corresponding to the power source is established to provide a basis for generating stable operation data; power demand forecast characteristics are obtained to obtain the changing demand of the grid load to provide a basis for dispatching decisions; environmental data is monitored, an environmental data set is established, and environmental information affecting power generation is obtained; the intelligent dispatching channel is initialized with power demand characteristics to prepare for equipment dispatching; the power generation parameters of clean equipment are predicted according to the environmental data set to provide a basis for generating stable operation data; equipment stable operation data is generated to ensure smooth dispatching; dispatching decisions are generated to form equipment dispatching instructions; and the technical solution for supply dispatch optimization of the power plant is completed with the dispatching decision results, which solves the technical problem that it is difficult to accurately predict and stably match clean power generation and grid load based on traditional models in the existing technology, resulting in low grid dispatching accuracy and poor coordination, and achieves the technical effect of real-time collaborative optimization of clean power generation and grid load and improving grid dispatching accuracy and coordination by building digital twins and intelligent dispatching channels.

[0009] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A flow chart of a scheduling algorithm optimization method based on power twin is provided for the embodiment of the present application;

[0011] Figure 2 A flow chart of establishing a prediction environment data set in a scheduling algorithm optimization method based on power twin is provided for an embodiment of the present application;

[0012] Figure 3 A structural schematic diagram of a scheduling algorithm optimization system based on power twin is provided for an embodiment of the present application.

[0013] Explanation of the accompanying drawings: twin model establishment module 11, evaluation function establishment module 12, demand feature acquisition module 13, environmental data monitoring module 14, channel initialization module 15, operation stability data module 16, scheduling decision result module 17, supply scheduling optimization module 18. DETAILED DESCRIPTION

[0014] The overall idea of ​​the technical solution provided by this application is as follows:

[0015] The embodiments of the present application provide a scheduling algorithm optimization method and system based on power twins, which actively adapts to the power generation output characteristics of various types of cleaning equipment and changes in grid load demand based on the constructed digital twin system and intelligent scheduling channel, thereby achieving coordinated optimization of the grid and the power generation end.

[0016] First, a digital twin model of the power plant is established to monitor environmental data, predict the power generation parameters of the clean equipment, and generate the equipment's operational stability data. Then, the intelligent dispatching channel is used to receive operational stability data, real-time environmental data sets, and predicted environmental data sets, and a dispatching decision is made on this basis. Finally, the dispatching decision is used to guide power supply dispatching.

[0017] Through this application, active prediction and tracking of the power generation characteristics of clean equipment are achieved. Through the coordination of intelligent scheduling channels, the overall scheduling level of the power grid for various types of clean power generation is effectively improved, the precise matching of the power grid and the power generation end is achieved, and the accuracy and coordination of power scheduling are greatly improved.

[0018] After introducing the basic principles of the present application, various non-limiting implementation methods of the present application will be specifically described below in conjunction with the drawings in the specification.

[0019] Embodiment 1

[0020] like Figure 1 As shown, the embodiment of the present application provides a scheduling algorithm optimization method based on power twins, the method comprising:

[0021] Establishing a digital twin model of the power plant, wherein the digital twin model is constructed by extracting equipment data of the power plant, wherein the equipment data includes equipment parameter data and equipment historical working data;

[0022] In an embodiment of the present application, the digital twin model of the power plant reflects the comprehensive information of the power plant and realizes the mapping from physical space to virtual space. First, the equipment data of the power plant is extracted from the equipment database of the power plant, including equipment parameter data and equipment historical working data. Among them, the equipment parameter data reflects the static information of the equipment, such as model parameters, rated parameters, etc.; the equipment historical working data reflects the dynamic information of the equipment, such as historical operating data such as operating power. Then, through physical modeling and mining and fusion technology of multi-source heterogeneous data, a digital twin model including a physical model and a data model is built to reflect the operating status, performance parameters and other information of the power plant in real time. Among them, the physical model is established using the theory of power plant equipment, such as the electromechanical energy conversion model; the data model is established by analyzing the statistical laws of historical data. Real-time reflection of the operating status, performance parameters and other information of the power plant.

[0023] Execute power source identification, and establish a device stability evaluation function corresponding to the power source through the digital twin model;

[0024] In the embodiment of the present application, after obtaining the digital twin model, it is necessary to clarify the power generation equipment corresponding to different power sources, that is, to perform power source identification. Among them, the power source refers to various types of renewable energy power generation devices in the power plant, such as wind turbines, photovoltaic power generation components, etc. First, in the digital twin model, identify different types of power generation equipment in the power plant and mark their power source attributes. For example, if a certain power source is a wind turbine, its power source identification is wind energy, and if a certain power source is a photovoltaic power generation component, its power source is solar energy.

[0025] Next, establish a stability evaluation function for the equipment corresponding to different power sources. This function is based on the output characteristics of the power generation equipment under specific environmental conditions and evaluates the operating stability of the equipment under such conditions. For example, the evaluation function of a wind turbine can be established based on the wind speed-power characteristic curve; the photovoltaic power generation component can use the light intensity-output characteristics to establish the equipment stability evaluation function.

[0026] Through the establishment of power source identification and equipment stability evaluation functions, a foundation has been laid for the subsequent generation of operation stability data and power generation scheduling decisions.

[0027] Acquiring power demand characteristics, wherein the power demand characteristics are prediction characteristics;

[0028] In an embodiment of the present application, first, the power load curve data of the power plant in the past 1-3 years is collected, and the time granularity is 1 hour to form the original load data. Secondly, the collected original load data is pre-processed by cleaning, filtering, missing value filling, etc., and the abnormal data is adjusted to normal values. Then, the BP neural network prediction model is adopted, and the load of the past 24 hours is selected as the model input, and the target output is the load after 1 hour. Afterwards, the first 70% of the data in the historical data is used as the training data set, and the last 30% of the data is used as the test set. The parameters are repeatedly trained and adjusted so that the prediction error of the test set meets the requirements, and the power demand prediction model is obtained. Subsequently, the trained power demand prediction model is used every 1 hour, the load data of the last 24 hours is input, the predicted load data after 1 hour is predicted, and the peak load, valley load and coefficient of variation in the predicted load data are extracted to obtain the power demand characteristics.

[0029] Monitor environmental data of clean power generation through monitoring sensors, and establish real-time environmental data sets and predicted environmental data sets;

[0030] In the embodiment of the present application, first, monitoring sensors such as anemometers, wind vanes, and light meters are arranged at the height and position corresponding to clean power generation equipment such as wind turbines and photovoltaic power generation arrays. Then, environmental data such as wind speed, wind direction, and light intensity collected by the monitoring sensors are collected in real time through wired or wireless networks such as RS485 buses. Next, the collected raw environmental data is pre-processed by filtering, smoothing, and normalization to generate a real-time environmental data set with good quality.

[0031] At the same time, the historical environmental data of the power plant in the past 2-3 years, including wind speed, wind direction, temperature, air pressure, solar radiation, etc., are collected at the hourly granularity, and the collected historical environmental data are preprocessed by missing value filling, noise filtering, error detection, etc. Then, LSTM, GRU neural network, etc. are selected as environmental prediction models, and the environmental prediction model is trained using the preprocessed historical environmental data to optimize the model parameters. Afterwards, given the past time series, the trained environmental prediction model is used to predict the environmental data within a time range of 1 hour from now on, and the predicted environmental data set is obtained.

[0032] Inputting the power demand characteristics into the intelligent dispatching channel to initialize the channel of the intelligent dispatching channel;

[0033] In an embodiment of the present application, first, the input and output variables of the intelligent dispatching channel are determined. The input variables include stable operation data, real-time environmental data sets, and predicted environmental data sets. The output variables are the start and stop and power output control instructions of the power generation equipment as the dispatching decision results. Then, a dispatching optimization model structure based on deep reinforcement learning is designed and constructed, and at the same time, reward mechanisms such as carbon emission reduction and economy are formulated to obtain an intelligent dispatching channel. Next, the acquired power demand characteristics are input into the constructed intelligent dispatching channel to complete the initialization of all model parameters, so that the power dispatching meets the power demand characteristics, so as to achieve targeted and highly adaptable power dispatching.

[0034] Predicting power generation parameters of the cleaning equipment according to the real-time environmental data and the predicted environmental data set, and generating operation stability data according to the prediction result and the equipment stability evaluation function;

[0035] In an embodiment of the present application, first, a mapping relationship model between environmental data and power generation parameters of cleaning equipment is established based on a historical environmental data set, such as a mapping relationship model between wind turbine output power and wind speed, and between photovoltaic array output and sunshine intensity. Then, the real-time environmental data and the predicted environmental data set are input into the mapping relationship model in sequence, and the predicted power generation parameters, including output power, electricity, etc., are calculated every 15 minutes for the next 24 hours to obtain the predicted results. Afterwards, a stable operation threshold is set as the maximum variation range allowed for clean power generation equipment in different environments, and exceeding it is an unstable state. Subsequently, the predicted results are input into the equipment stability evaluation function to obtain the expected stability evaluation value of the cleaning equipment at each moment in the future. The expected stability evaluation value is compared with the preset stable operation threshold. If it is lower than or equal to the threshold, it is considered to be a stable state. Afterwards, the stable state judgment results of all predicted times are sorted out to form the future operation stability data of the equipment.

[0036] Inputting the operation stability data, the real-time environment data set and the predicted environment data set into the intelligent scheduling channel to generate a scheduling decision result;

[0037] In an embodiment of the present application, after obtaining the operational stability data, the real-time environmental data set and the predicted environmental data set, the operational stability data, the real-time environmental data set and the predicted environmental data set are integrated and sent to the intelligent scheduling channel, which comprehensively judges the environmental resource status, power load forecast, and equipment stability, and generates the optimal power generation output scheduling plan for clean equipment such as wind turbines and photovoltaic arrays in the future scheduling cycle as a scheduling decision result.

[0038] The supply scheduling optimization of the power plant is completed according to the scheduling decision results.

[0039] In the embodiment of the present application, the obtained scheduling decision results determine the optimal start-stop and power output control strategy of the clean power generation equipment in each time period. According to the scheduling decision results, scheduling instructions are issued to wind turbines, photovoltaic arrays and other equipment. For example, wind turbines 1 and 2 are shut down in time period 1, and photovoltaic arrays 1 and 3 are issued a start control instruction in time period 2. After receiving the instruction, the equipment adjusts its own operation and maintenance status or output strategy accordingly, thereby achieving coordinated and efficient operation of clean power generation equipment in the power plant and optimizing the supply scheduling of the power generation end.

[0040] Furthermore, the embodiment of the present application also includes:

[0041] Establishing a supervisory response model, wherein the supervisory response model is a supervisory response model for performing real-time monitoring;

[0042] When any scheduling decision result is generated, a supervisory response instruction is generated synchronously;

[0043] Controlling the monitoring sensor to perform power supply data monitoring of the power plant through the supervisory response instruction to generate a monitoring data set;

[0044] Synchronously inputting the monitoring data set and the corresponding scheduling decision result into the supervisory response model to generate a supervisory response reward result;

[0045] Subsequent scheduling decision optimization is performed based on the supervisory response reward result.

[0046] In a feasible implementation, first, collect historical dispatch decision data and corresponding power plant operation monitoring data, including environmental data, equipment output data, etc., remove erroneous data, standardize the data format, annotate data attributes, and construct a sample set. Then, select the RNN structure, set the input layer, multi-layer LSTM layer and output layer, etc., use the supervised learning method to train with the collected data, and establish a mapping relationship between the monitoring data set, the corresponding dispatch decision results and the supervisory response reward results. Next, test the model effect, evaluate the accuracy of the supervisory response model, adjust the network structure and parameters, repeat the training and evaluation, and obtain a trained supervisory response model.

[0047] When the intelligent dispatch channel outputs the dispatch decision result, that is, when the dispatch decision result at a certain moment is obtained, a supervisory response instruction is generated synchronously to drive the monitoring of power supply data. After receiving the supervisory response instruction, the power monitoring sensors configured in various instruments in the power plant are controlled to collect the operation data of the designated device according to the requirements of the supervisory response instruction, such as measuring and recording the power supply parameters such as active power, voltage, and current of wind turbines, photovoltaic panels and other equipment. After preprocessing, a monitoring data set for a specific dispatch decision result is summarized to record the actual operation of the equipment under the guidance of the dispatch decision result.

[0048] Subsequently, the obtained monitoring data set and the corresponding dispatch decision results are synchronously input into the pre-trained supervisory response model, which analyzes and compares the two, evaluates the effect of the dispatch decision, and judges its quality. For example, the supervisory response model identifies whether the dispatch decision result causes the photovoltaic power generation to exceed the environmental capacity, or whether the wind turbine fails and stops, etc., thereby generating a numerical scoring result as a supervisory response reward result, indicating the quality of the dispatch decision result. The higher the supervisory response reward result, the better the decision quality; the negative value or the lower the value, the worse the dispatch decision effect, and the greater the room for improvement of the dispatch decision result. Afterwards, the supervisory response reward result is input into the intelligent dispatch channel, and the state behavior space and reward mechanism are gradually adjusted, in order to generate better decisions in the new environment, realize the subsequent dispatch decision optimization, and thus continuously improve the overall benefits of clean equipment power generation dispatch.

[0049] Furthermore, the embodiment of the present application also includes:

[0050] Establishing a continuous compensation coefficient for monitoring response reward results;

[0051] Determine the reinforcement signal of the reward end according to the continuous compensation coefficient and the supervisory response reward result, and generate a first reinforcement signal determination result;

[0052] Analyze the scheduling decision result, and generate a second enhanced signal determination result according to the analysis result;

[0053] determining a reinforcement node based on the first reinforcement signal determination result and the second reinforcement signal determination result;

[0054] Subsequent scheduling decision optimization is performed based on the enhanced nodes.

[0055] In a feasible implementation, considering the dynamic characteristics of the power dispatching environment, the one-time reward evaluation results that only rely on the supervisory response reward results may lead to the fragmentation of dispatching decisions. In order to make the dispatching decision optimization smoother and more coherent, by analyzing the time decay law of the historical supervisory response reward results, the supervisory response reward results at the current moment are used to calculate the correlation weights in the same period of history and in the future, and form a continuous compensation coefficient for the supervisory response reward results. Then, based on the generated continuous compensation coefficient, combined with the supervisory response reward results at the current moment, the supervisory response reward results after compensation adjustment are calculated, and the continuity of the adjusted supervisory response reward results in the recent time period is judged, thereby generating the first enhanced signal judgment result. Afterwards, the current dispatching decision results are analyzed to determine whether the start and stop adjustment range of the equipment in the results exceeds the stable range, and the comprehensive benefit value of the dispatching strategy to economy and stability is analyzed, thereby generating the second enhanced signal judgment result.

[0056] Then, based on the first reinforcement signal determination result and the second reinforcement signal determination result, a comprehensive determination is made. If the first reinforcement signal determination result and the second reinforcement signal determination result are activated at the same time, the current moment is determined to be the reinforcement node for the subsequent start of reinforcement learning. Then, starting from the confirmed reinforcement node, the online reinforcement learning process of the intelligent scheduling channel is triggered, and the scheduling data, environmental data, reward feedback, etc. in the recent period are revisited to find a better strategy for subsequent scheduling decisions.

[0057] Further, such as Figure 2 As shown, the embodiment of the present application also includes:

[0058] Acquire a periodic environment data set, and perform a temporal environment complexity evaluation on the periodic environment data set to establish a temporal complexity coefficient;

[0059] Performing window size constraint on the prediction window by using the timing complexity coefficient;

[0060] Perform data collection through the monitoring sensor at any time series node, and use the collection result as the real-time environment data set of the current time series node;

[0061] The prediction window is matched with the current time series node, and the window environment change prediction is performed based on the matching result with the time series node trend as the prediction reference, and the prediction environment data set is established based on the prediction result.

[0062] In a feasible implementation, considering that the power dispatching environment has the characteristics of periodic changes, in order to dynamically set the time range of environmental prediction, first, collect historical periodic environmental data sets, such as daily wind speed and light data sets in the past year. Then, use the time series complexity analysis algorithm to evaluate the complexity of the periodic environmental data set in different period segments, such as the complexity of the working day cycle and the holiday cycle, so as to determine the time series complexity coefficient representing the complexity and unification of the periodic time series environment, so as to guide the subsequent adaptive constraints and settings of the size of the environmental prediction time window. Afterwards, according to the time series complexity coefficient, the prediction window range of the environment is adaptively constrained. If the complexity of the current environmental cycle segment is high, the backward time span of the prediction window is set to be shorter, such as 6 hours; if the time series complexity coefficient is low, the prediction window range is expanded, such as 24 hours. By dynamically adjusting the size of the prediction time window, the accuracy of the prediction is guaranteed, and the error diffusion problem caused by the over-long window is avoided.

[0063] After setting the window size constraint of the prediction window, at each time point, the data monitoring sensors such as wind speed, temperature and humidity arranged on site are used to collect and record the environmental parameters at that time point. For example, at the time series node of 8 o'clock in the morning, the monitoring instrument is triggered to detect the on-site environment of the wind farm, photovoltaic power station, etc., and a data set representing the current environmental state of the time series node is generated, that is, the real-time environmental data set. Subsequently, the time series node is used as the starting time, and the set prediction time window size is matched thereafter. For example, if the current time series point is 1 o'clock in the afternoon, the matching 6-hour window will be located in the time period from 1 o'clock to 7 o'clock to obtain the matching result. Next, the data set of the window period in the historical environment database is loaded, the data change trend of the historical period is analyzed, and the current data of 1 o'clock is used as the reference starting point to predict the values ​​of the environmental parameters in the window from 1 o'clock to 7 o'clock, thereby obtaining the predicted environmental data set.

[0064] Furthermore, the embodiment of the present application also includes:

[0065] Establishing a window environment threshold corresponding to the prediction window;

[0066] Conduct trigger evaluation of the sequence window environment threshold on the real-time environment data set to generate trigger results;

[0067] Taking the trigger result as the search center, performing a preceding trend verification of the search center;

[0068] If the previous trend verification result is consistent with the trigger result, a dynamic window adjustment instruction is generated;

[0069] The prediction window is adaptively updated through the dynamic window adjustment instruction, and the prediction environment data set is established through the adaptively updated prediction window.

[0070] In a feasible implementation, taking into account the variability of the power dispatching environment, in order to achieve flexible adjustment of the prediction window, based on the distribution of historical data in each prediction period, the reasonable maximum and minimum thresholds of the environmental parameters in the period are determined to obtain the window environmental threshold corresponding to the prediction window. For example, for the prediction window from 10 a.m. to 4 a.m. on the same day, the upper and lower limits of the wind speed range threshold for normal operation of the unit are analyzed based on the historical data of the same period. Then, according to the window environmental threshold corresponding to the established prediction window, the latest real-time environmental data set collected is judged to see whether its parameter value exceeds the sequence window environmental threshold, thereby obtaining a trigger result. Among them, the sequence window environmental threshold is the environmental threshold range of a prediction window after the prediction window where the current time series node is located.

[0071] Subsequently, the trigger result is used as the search center to search forward for the change trend of environmental data within a period of time. For example, if the wind speed at 10:05 am of the current time series node triggers the threshold judgment of the next window, the wind speed data set of 10-30 minutes before and after 10:05 is balanced to analyze and determine whether the wind speed trend in this period is also over-limit. By performing the pre-trend verification of the search center, false alarms caused by accidental abnormal data can be avoided, thereby enhancing the robustness of the basis for adjusting the prediction window. If the pre-trend verification result is consistent with the trigger result, it means that this is not an accidental temporary anomaly. At this time, the fact that the environment has suddenly changed is confirmed, and a dynamic window adjustment instruction for adjusting the prediction window is generated to start the dynamic change of the prediction window. Then, the adaptive update process of the environmental prediction mechanism is started through the dynamic window adjustment instruction. According to the real-time state and change judgment of the environment, the subsequent prediction time period of the current time series node is reset to form an updated reasonable window size. For example, the original 10 o'clock-4 o'clock prediction window is adjusted to 10 o'clock to 12 o'clock. After obtaining the adjusted forecast window, the previous environmental forecast process is repeated to generate a matching forecast environment data set based on the new forecast window to adapt to the dynamic changes in the power dispatch environment.

[0072] Furthermore, the embodiment of the present application also includes:

[0073] Analyzing the power demand characteristics to generate a time-series steady-state value of the power demand characteristics;

[0074] An accurate evaluation is performed on the predicted power demand characteristics, a stability influence coefficient is generated according to the evaluation result, and a time series steady-state value compensation is performed through the stability influence coefficient;

[0075] Generate decision preference constraints based on compensated time series steady-state values;

[0076] The decision preference constraint is used to perform initialization constraints on the intelligent scheduling channel.

[0077] In a feasible implementation, first, analyze the various characteristics reflected in the obtained power demand characteristics, determine the stability requirements of the load curves in different time periods on the power supply dispatch, and extract the time series steady-state value as the balance of power supply and demand in the time period. For example, the benchmark requirement value for the stability of the power dispatch strategy during the peak period in the morning and afternoon of weekdays is higher than that during the night period. Then, further evaluate the accuracy and reliability of the power demand characteristics themselves to reasonably compensate for the time series steady-state value. Through methods such as sampling error testing, the confidence and accuracy of the power load forecast in different time periods in the power demand characteristics are determined, so as to set the stability influence coefficient corresponding to the time period. For example, the time period with a large prediction error has a higher influence coefficient. Subsequently, based on the accuracy influence coefficient, the time series steady-state value of each time period is compensated and adjusted to obtain the time series steady-state value after accuracy compensation.

[0078] Afterwards, the constraints of the scheduling decision preference in different time periods are determined based on the compensated time series steady-state values. For example, the compensated stability value during the peak hours of weekdays with high stability requirements is set as a constraint to keep the decision preference stable; the low-peak hours at night are set as constraints that allow the corresponding adjustment range, forming a series of decision preference constraints that distinguish time and goals. After that, the set decision preference constraints are loaded into the intelligent scheduling channel, and the intelligent scheduling channel is initialized as a constraint item of the intelligent scheduling channel. When the intelligent scheduling channel generates scheduling decision results, it must comply with this set of decision preference constraints, so that the scheduling decisions within the forecast period are more reasonable and stable.

[0079] Furthermore, the embodiment of the present application also includes:

[0080] Set deviation warning thresholds;

[0081] Determine whether the difference between the monitoring data set and the corresponding scheduling decision result meets the deviation warning threshold;

[0082] If the conditions are met, the device abnormality is reported, and early warning management of the device operation is performed through the device abnormality.

[0083] In a feasible implementation, considering the complexity of equipment operation and maintenance, in order to realize fault warning of clean power generation equipment such as wind turbines and photovoltaic modules, firstly, according to the historical statistical analysis of equipment operation and maintenance, the error range between the monitoring status data and the corresponding dispatch decision results of different types of equipment under normal working conditions is determined. On this basis, under the premise of ensuring a certain tolerance space, a deviation warning threshold is set, such as setting the deviation between the monitoring status data and the corresponding dispatch decision results to no more than 8%, which is the basis for judging whether the equipment is abnormal.

[0084] Then, the difference between the most recent dispatch decision result and the corresponding equipment actual monitoring data set is compared to determine whether the deviation between the two is within the required range of the deviation warning threshold. Taking the dispatch decision result of "PV module A increases the load by 100 kilowatts" as an example, the actual power monitoring curve of PV module A before and after the decision is calculated, and the deviation between the monitoring curve and the rated value of 100 kilowatts is analyzed. If the deviation is within the 8% error range, it is judged to be normal and does not meet the deviation warning threshold; if the deviation exceeds the 8% error range, it meets the deviation warning threshold. At this time, the abnormal judgment is confirmed, and the monitoring warning information of the abnormal equipment deviation is formally reported, and the online status monitoring and maintenance process of the equipment deviation is automatically started. The relevant operation and maintenance personnel quickly check the equipment, analyze the root cause, and handle it according to the regulations, correct the deviation, and restore the equipment to normal working state to realize the early warning management of equipment operation.

[0085] In summary, the scheduling algorithm optimization method based on power twin provided in the embodiment of the present application has the following technical effects:

[0086] Establish a digital twin model of the power plant to provide a basis for building equipment stability evaluation functions and power generation predictions. Perform power source identification, and establish equipment stability evaluation functions corresponding to power sources through digital twin models to distinguish different power sources and support scheduling optimization. Obtain power demand characteristics to provide a basis for scheduling decisions. Use monitoring sensors to monitor environmental data of clean power generation, establish real-time environmental data sets and predicted environmental data sets, and establish environmental influencing factor data sets. Input power demand characteristics into the intelligent scheduling channel to initialize the channel of the intelligent scheduling channel and prepare for equipment coordinated scheduling. According to the real-time environmental data and predicted environmental data sets, predict the power generation parameters of clean equipment, and generate operation stability data according to the prediction results and equipment stability evaluation function to ensure smooth scheduling. Input operation stability data, real-time environmental data sets, and predicted environmental data sets into the intelligent scheduling channel to generate scheduling decision results and form equipment scheduling decision plans. According to the scheduling decision results, complete the supply scheduling optimization of the power plant, guide the power plant to optimize operation according to the scheduling decision, and improve the accuracy and coordination of power grid scheduling. Example

[0087] Based on the same inventive concept as the scheduling algorithm optimization method based on power twin in the aforementioned embodiment, Figure 3 As shown, the embodiment of the present application provides a scheduling algorithm optimization system based on power twins, the system comprising:

[0088] A twin model building module 11 is used to build a digital twin model of a power plant, wherein the digital twin model is built by extracting equipment data of the power plant, wherein the equipment data includes equipment parameter data and equipment historical working data;

[0089] An evaluation function establishment module 12, used to perform power source identification and establish a device stability evaluation function corresponding to the power source through the digital twin model;

[0090] The demand feature acquisition module 13 is used to acquire the power demand feature, wherein the power demand feature is a prediction feature;

[0091] An environmental data monitoring module 14 is used to monitor environmental data of clean power generation through monitoring sensors, and to establish a real-time environmental data set and a predicted environmental data set;

[0092] A channel initialization module 15 is used to input the power demand characteristics into the intelligent scheduling channel to initialize the channel of the intelligent scheduling channel;

[0093] An operation stability data module 16 is used to predict power generation parameters of the cleaning equipment according to the real-time environmental data and the predicted environmental data set, and generate operation stability data according to the prediction result and the equipment stability evaluation function;

[0094] The scheduling decision result module 17 is used to input the operation stability data, the real-time environment data set and the predicted environment data set into the intelligent scheduling channel to generate a scheduling decision result;

[0095] The supply scheduling optimization module 18 is used to optimize the supply scheduling of the power plant according to the scheduling decision result.

[0096] Furthermore, the embodiment of the present application also includes a scheduling decision optimization module, which includes the following execution steps:

[0097] Establishing a supervisory response model, wherein the supervisory response model is a supervisory response model for performing real-time monitoring;

[0098] When any scheduling decision result is generated, a supervisory response instruction is generated synchronously;

[0099] Controlling the monitoring sensor to perform power supply data monitoring of the power plant through the supervisory response instruction to generate a monitoring data set;

[0100] Synchronously inputting the monitoring data set and the corresponding scheduling decision result into the supervisory response model to generate a supervisory response reward result;

[0101] Subsequent scheduling decision optimization is performed based on the supervisory response reward result.

[0102] Furthermore, the scheduling decision optimization module also includes the following execution steps:

[0103] Establishing a continuous compensation coefficient for monitoring response reward results;

[0104] Determine the reinforcement signal of the reward end according to the continuous compensation coefficient and the supervisory response reward result, and generate a first reinforcement signal determination result;

[0105] Analyze the scheduling decision result, and generate a second enhanced signal determination result according to the analysis result;

[0106] determining a reinforcement node based on the first reinforcement signal determination result and the second reinforcement signal determination result;

[0107] Subsequent scheduling decision optimization is performed based on the enhanced nodes.

[0108] Furthermore, the embodiment of the present application also includes an environmental data prediction module, which includes the following execution steps:

[0109] Acquire a periodic environment data set, and perform a temporal environment complexity evaluation on the periodic environment data set to establish a temporal complexity coefficient;

[0110] Performing window size constraint on the prediction window by using the timing complexity coefficient;

[0111] Perform data collection through the monitoring sensor at any time series node, and use the collection result as the real-time environment data set of the current time series node;

[0112] The prediction window is matched with the current time series node, and the window environment change prediction is performed based on the matching result with the time series node trend as the prediction reference, and the prediction environment data set is established based on the prediction result.

[0113] Furthermore, the embodiment of the present application also includes a prediction window update module, which includes the following execution steps:

[0114] Establishing a window environment threshold corresponding to the prediction window;

[0115] Conduct trigger evaluation of the sequence window environment threshold on the real-time environment data set to generate trigger results;

[0116] Taking the trigger result as the search center, performing a preceding trend verification of the search center;

[0117] If the previous trend verification result is consistent with the trigger result, a dynamic window adjustment instruction is generated;

[0118] The prediction window is adaptively updated through the dynamic window adjustment instruction, and the prediction environment data set is established through the adaptively updated prediction window.

[0119] Furthermore, the embodiment of the present application also includes a channel initialization constraint module, which includes the following execution steps:

[0120] Analyzing the power demand characteristics to generate a time-series steady-state value of the power demand characteristics;

[0121] An accurate evaluation is performed on the predicted power demand characteristics, a stability influence coefficient is generated according to the evaluation result, and a time series steady-state value compensation is performed through the stability influence coefficient;

[0122] Generate decision preference constraints based on compensated time series steady-state values;

[0123] The decision preference constraint is used to perform initialization constraints on the intelligent scheduling channel.

[0124] Furthermore, the embodiment of the present application also includes a device early warning management module, which includes the following execution steps:

[0125] Set deviation warning thresholds;

[0126] Determine whether the difference between the monitoring data set and the corresponding scheduling decision result meets the deviation warning threshold;

[0127] If the conditions are met, the device abnormality is reported, and early warning management of the device operation is performed through the device abnormality.

[0128] Any step of the method described above can be stored as a computer instruction or program in an unlimited computer memory, and can be called and recognized by an unlimited computer processor to implement any method in the embodiments of the present application, without any unnecessary restrictions.

[0129] Furthermore, the first or second mentioned above may not only represent an order relationship, but may also represent a specific concept, and / or refer to multiple elements that can be selected individually or in full. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. The scheduling algorithm optimization method based on power twin is characterized by: The method comprises: Establishing a digital twin model of the power plant, wherein the digital twin model is constructed by extracting equipment data of the power plant, wherein the equipment data includes equipment parameter data and equipment historical working data; Execute power source identification, and establish a device stability evaluation function corresponding to the power source through the digital twin model; Acquiring power demand characteristics, wherein the power demand characteristics are prediction characteristics; Monitor environmental data of clean power generation through monitoring sensors, and establish real-time environmental data sets and predicted environmental data sets; Inputting the power demand characteristics into the intelligent dispatching channel to initialize the channel of the intelligent dispatching channel; Predicting power generation parameters of the cleaning equipment according to the real-time environmental data and the predicted environmental data set, and generating operation stability data according to the prediction result and the equipment stability evaluation function; Inputting the operation stability data, the real-time environment data set and the predicted environment data set into the intelligent scheduling channel to generate a scheduling decision result; completing supply scheduling optimization of the power plant according to the scheduling decision result; The method further comprises: Establishing a supervisory response model, wherein the supervisory response model is a supervisory response model for performing real-time monitoring; When any scheduling decision result is generated, a supervisory response instruction is generated synchronously; Controlling the monitoring sensor to perform power supply data monitoring of the power plant through the supervisory response instruction to generate a monitoring data set; Synchronously inputting the monitoring data set and the corresponding scheduling decision result into the supervisory response model to generate a supervisory response reward result; Subsequent scheduling decision optimization is performed based on the supervisory response reward result.

2. The method according to claim 1, characterized in that The method further comprises: Establishing a continuous compensation coefficient for monitoring response reward results; Determine the reinforcement signal of the reward end according to the continuous compensation coefficient and the supervisory response reward result, and generate a first reinforcement signal determination result; Analyze the scheduling decision result, and generate a second enhanced signal determination result according to the analysis result; determining a reinforcement node based on the first reinforcement signal determination result and the second reinforcement signal determination result; Subsequent scheduling decision optimization is performed based on the enhanced nodes.

3. The method according to claim 1, characterized in that The method further comprises: Acquire a periodic environment data set, and perform a temporal environment complexity evaluation on the periodic environment data set to establish a temporal complexity coefficient; Performing window size constraint on the prediction window by using the timing complexity coefficient; Perform data collection through the monitoring sensor at any time series node, and use the collection result as the real-time environment data set of the current time series node; The prediction window is matched with the current time series node, and the window environment change prediction is performed based on the matching result with the time series node trend as the prediction reference, and the prediction environment data set is established based on the prediction result.

4. The method according to claim 3, characterized in that The method further comprises: Establishing a window environment threshold corresponding to the prediction window; Conduct trigger evaluation of the sequence window environment threshold on the real-time environment data set to generate trigger results; Taking the trigger result as the search center, performing a preceding trend verification of the search center; If the previous trend verification result is consistent with the trigger result, a dynamic window adjustment instruction is generated; The prediction window is adaptively updated through the dynamic window adjustment instruction, and the prediction environment data set is established through the adaptively updated prediction window.

5. The method according to claim 1, characterized in that The method further comprises: Analyzing the power demand characteristics to generate a time-series steady-state value of the power demand characteristics; An accurate evaluation is performed on the predicted power demand characteristics, a stability influence coefficient is generated according to the evaluation result, and a time series steady-state value compensation is performed through the stability influence coefficient; Generate decision preference constraints based on compensated time series steady-state values; The decision preference constraint is used to perform initialization constraints on the intelligent scheduling channel.

6. The method according to claim 1, characterized in that The method further comprises: Set deviation warning thresholds; Determine whether the difference between the monitoring data set and the corresponding scheduling decision result meets the deviation warning threshold; If the conditions are met, the device abnormality is reported, and early warning management of the device operation is performed through the device abnormality.

7. The dispatch algorithm optimization system based on power twin is characterized by: For implementing the dispatch algorithm optimization method based on power twin according to any one of claims 1 to 6, the system comprises: A twin model building module, wherein the twin model building module is used to build a digital twin model of a power plant, wherein the digital twin model is built by extracting equipment data of the power plant, wherein the equipment data includes equipment parameter data and equipment historical working data; An evaluation function establishment module, the evaluation function establishment module is used to perform power source identification and establish a device stability evaluation function corresponding to the power source through the digital twin model; A demand feature acquisition module, the demand feature acquisition module is used to acquire power demand features, wherein the power demand features are prediction features; An environmental data monitoring module, which is used to monitor environmental data of clean power generation through monitoring sensors, and to establish a real-time environmental data set and a predicted environmental data set; A channel initialization module, wherein the channel initialization module is used to input the power demand characteristics into the intelligent scheduling channel to perform channel initialization of the intelligent scheduling channel; An operation stability data module, the operation stability data module is used to predict the power generation parameters of the cleaning equipment according to the real-time environmental data and the predicted environmental data set, and generate operation stability data according to the prediction result and the equipment stability evaluation function; A scheduling decision result module, wherein the scheduling decision result module is used to input the operation stability data, the real-time environment data set and the predicted environment data set into the intelligent scheduling channel to generate a scheduling decision result; A supply scheduling optimization module, the supply scheduling optimization module is used to complete the supply scheduling optimization of the power plant according to the scheduling decision result; It also includes a scheduling decision optimization module, which performs the following steps: Establishing a supervisory response model, wherein the supervisory response model is a supervisory response model for performing real-time monitoring; When any scheduling decision result is generated, a supervisory response instruction is generated synchronously; Controlling the monitoring sensor to perform power supply data monitoring of the power plant through the supervisory response instruction to generate a monitoring data set; Synchronously inputting the monitoring data set and the corresponding scheduling decision result into the supervisory response model to generate a supervisory response reward result; Subsequent scheduling decision optimization is performed based on the supervisory response reward result.

Images