Dynamic adjustment method and system for steam extraction energy storage and heat release in thermodynamic system

Abstract

The invention provides a dynamic adjustment method and system for steam extraction energy storage and heat release in a thermodynamic system, and relates to the technical field of energy consumption adjustment, and the method comprises the steps: obtaining multi-dimensional operation parameters of the thermodynamic system and heat consumption demand data of a user side, predicting a steam load state in a specific time period in the future, and dynamically dividing a load time period according to the predicted steam load state; on the basis of the divided load time periods, steam extraction energy storage and heat release operation is carried out in the corresponding load time periods according to the optimal operation strategies of all the time periods; and continuously optimizing the load prediction model according to the deviation data, based on the optimized load prediction model, integrating the real-time operation state of the thermodynamic system, the actual heat consumption condition of the user side and the energy market price information, constructing a multi-target optimization model, and adaptively adjusting the operation parameters of steam extraction energy storage and heat release by using the multi-target optimization model. And dynamic adjustment of steam extraction energy storage and heat release in the thermodynamic system is realized.

Description

Technical Field

[0001] The invention relates to energy consumption regulation technology, and in particular to a method and system for dynamically regulating steam extraction energy storage and heat release in a thermal system. Background Art

[0002] The existing thermal system operation and scheduling mainly rely on manual experience and fixed operation strategies, which are difficult to adapt to load fluctuations and diverse user needs. Traditional load forecasting methods are not accurate and cannot accurately grasp the trend of load changes. The scheduling optimization of thermal systems mainly focuses on a single goal, such as minimizing costs or maximizing efficiency, and lacks consideration of multi-objective collaborative optimization. The system operation lacks real-time monitoring and feedback mechanisms, and cannot dynamically adjust operating parameters, making it difficult to cope with real-time changes in load. Summary of the invention

[0003] The embodiments of the present invention provide a method and system for dynamically regulating steam extraction energy storage and heat release in a thermal system, which can solve the problems in the prior art.

[0004] According to a first aspect of the embodiments of the present invention,

[0005] A method for dynamically regulating steam extraction energy storage and heat release in a thermal system is provided, comprising:

[0006] Acquire multi-dimensional operating parameters of the thermal system and heat demand data of the user end, wherein the heat demand data includes user type, historical heat demand curve and real-time heat consumption, analyze the multi-dimensional operating parameters and heat demand data based on the deep learning algorithm, build a thermal system load prediction model, use the thermal system load prediction model, combine weather forecast data and user behavior patterns, predict the steam load state in a specific period of time in the future, and dynamically divide the load period into peak period, flat period and valley period according to the predicted steam load state;

[0007] Based on the divided load periods, a multi-energy collaborative optimization model is established. By solving the multi-energy collaborative optimization model, the optimal operation strategy for each period is obtained. According to the optimal operation strategy for each period, steam extraction storage and heat release operations are performed in the corresponding load period.

[0008] During the steam extraction storage and heat release process, a closed-loop feedback mechanism is established to compare the deviation between the predicted steam load and the actual load in real time, and dynamically fine-tune the steam extraction, storage and heat release. At the same time, the load prediction model is continuously optimized according to the deviation data. Based on the optimized load prediction model, the real-time operating status of the thermal system, the actual heat usage at the user end and the energy market price information are integrated to construct a multi-objective optimization model. The multi-objective optimization model is used to adaptively adjust the operating parameters of steam extraction storage and heat release, so as to realize dynamic regulation of steam extraction storage and heat release in the thermal system.

[0009] In an optional embodiment,

[0010] Based on deep learning algorithms, we analyze multi-dimensional operating parameters and heat demand data to build a thermal system load prediction model. We use the thermal system load prediction model, combined with weather forecast data and user behavior patterns, to predict the steam load state in a specific period of time in the future. Based on the predicted steam load state, we dynamically divide the load period into peak period, off-peak period and valley period.

[0011] Preprocess the multi-dimensional operating parameters of the thermal system and the heat demand data of the user end to generate a standardized multi-source data set, extract knowledge elements from the standardized multi-source data set using natural language processing technology, and construct a multi-level knowledge graph including the equipment layer, system layer and application layer;

[0012] Based on the multi-level knowledge graph, the thermal system is abstracted into a dynamic weighted graph network, in which nodes represent system components and edges represent physical connections and energy flows between components. A hierarchical recurrent neural network is constructed for the multi-scale time series data in the standardized multi-source data set, and the semantic information in the multi-level knowledge graph is used as additional input to extract multi-scale time series features.

[0013] Design a multimodal deep fusion network to adaptively fuse system topology features, timing features, and semantic features to build a load forecasting model. Use a multi-scale convolutional neural network to extract features from user heat data. Combine user attribute information in a multi-level knowledge graph, and dynamically fuse the extracted features through an attention mechanism to build a user behavior model.

[0014] The load forecasting basic model, user behavior model and anomaly detection model are integrated into a multi-task learning framework, and combined with the semantic enhancement features to generate multi-time scale load forecasting sequences. The load forecasting sequences are divided into load periods, including peak period, off-peak period and valley period, using an adaptive load partitioning algorithm.

[0015] In an optional embodiment,

[0016] The load forecast sequence is divided into the following steps using an adaptive load division algorithm, and the load period division results generated include:

[0017] The load forecasting sequence is subjected to wavelet transformation to obtain multi-layer scale coefficients and fluctuation coefficients. Adaptive weights are introduced to perform weighted fusion on the obtained multi-layer scale coefficients and fluctuation coefficients to obtain a feature sequence. The obtained feature sequence is used as a sample input into a pre-built process mixture model for clustering to obtain clustering results and cluster centers.

[0018] The typical load pattern with the highest similarity to the cluster center is retrieved from the pre-constructed multi-level knowledge graph, and the similarity between the cluster center and the predefined load pattern is calculated by cosine similarity. For each cluster, the predefined load pattern with the highest similarity is selected as the semantic label. Based on the obtained clustering results and semantic labels, a time period division result is generated for each moment in the load sequence, and according to the differences in load characteristics in different time periods, the generated time period division result is fine-tuned to obtain the final load time period division result.

[0019] In an optional embodiment,

[0020] Based on the divided load periods, a multi-energy collaborative optimization model is established. By solving the multi-energy collaborative optimization model, the optimal operation strategy for each period is obtained. According to the optimal operation strategy for each period, steam extraction storage and heat release operations are performed in the corresponding load period, including:

[0021] According to the peak load period obtained, the steam supply gap at each moment is predicted, and a peak heat release optimization model is established with the goal of minimizing the user-side heat supply gap. The peak heat release optimization model is solved using the model predictive control algorithm to obtain the optimal heat release power at each moment, generate peak heat release instructions, control the energy storage device to release stored energy, and supplement the peak steam supply gap;

[0022] Monitor the load fluctuation during the divided off-peak load period. When the actual load deviates from the predicted load by more than the set threshold, the steam extraction or heat release regulation mechanism is triggered, and the control algorithm is used to calculate the regulation amount to smooth the load fluctuation.

[0023] According to the divided off-peak load period, the steam extraction amount at each moment is predicted, and an optimization model for off-peak steam extraction with the goal of maximizing the energy storage is established. The dynamic programming algorithm is used to solve the off-peak steam extraction optimization model, and the optimal steam extraction power at each moment is obtained. The off-peak steam extraction command is generated, and the steam extraction device is controlled to store the remaining steam in the energy storage device.

[0024] The price signals of electricity and gas market are introduced to build a multi-energy collaborative optimization model. The mixed integer nonlinear programming algorithm is used to solve the multi-energy collaborative optimization model to obtain the optimal day-ahead dispatch plan for electricity, gas and heat. Based on the optimal day-ahead dispatch plan obtained, the real-time electricity price and gas price information is used, and a rolling time domain optimization algorithm is used to update the load forecast and market price in each dispatching period to obtain the optimal operation strategy for each period. According to the optimal operation strategy for each period, steam extraction storage and heat release operations are performed in the corresponding load period.

[0025] In an optional embodiment,

[0026] Introduce electricity and gas market price signals, build a multi-energy collaborative optimization model, use mixed integer nonlinear programming algorithm to solve the multi-energy collaborative optimization model, and obtain the optimal day-ahead dispatch plan for electricity, gas, and heat. Based on the optimal day-ahead dispatch plan, use real-time electricity and gas price information, and use a rolling time domain optimization algorithm to update load forecasts and market prices in each dispatch period to obtain the optimal operation strategy for each period. According to the optimal operation strategy for each period, steam extraction storage and heat release operations are performed in the corresponding load period, including:

[0027] Construct a multi-energy collaborative optimization model, which takes into account the coupling characteristics of the thermal system, the power system and the gas system, and uses the concept of energy hub to abstract the three systems into a multi-energy flow network to describe the conversion and transmission relationship between various energy forms;

[0028] Introducing price signals from the electricity and gas markets, building a multi-objective optimization model, the objective functions of which include minimizing operating costs, maximizing energy efficiency, and optimizing environmental benefits;

[0029] Establish constraints for multi-energy collaborative optimization, including power system constraints, gas system constraints, thermal system constraints, energy hub constraints, and electric-thermal coupling constraints;

[0030] A mixed integer nonlinear programming algorithm is used to solve the multi-energy collaborative optimization model, wherein the problem is decomposed into a main problem and sub-problems using a Benders decomposition algorithm, and an optimal day-ahead dispatch plan for the power system, the gas system, and the thermal system is generated through iterative solution;

[0031] At the beginning of each scheduling period, the latest electricity price and gas price information is obtained, and the load forecast results are updated. Taking the current time as the starting point and the predetermined time window as the optimization range, a rolling optimization model is constructed, and the executed scheduling decisions are incorporated into the rolling optimization model as known conditions. The rolling optimization model is solved using a rolling time domain optimization algorithm to obtain the optimal scheduling strategy in the future time window, including the output adjustment and energy flow distribution of the power system, gas system and thermal system. According to the optimal scheduling strategy, steam extraction storage and heat release operations are performed in the corresponding load period.

[0032] In an optional embodiment,

[0033] During the steam extraction and heat release process, a closed-loop feedback mechanism is established to compare the deviation between the predicted steam load and the actual load in real time, and dynamically adjust the steam extraction, energy storage and heat release, including:

[0034] Establish a high-precision dynamic model of the steam extraction energy storage and heat release system, which takes the steam extraction amount, energy storage amount and heat release amount as input variables, and takes system efficiency, energy consumption and steam output as output variables. Based on the established high-precision dynamic model, design a model predictive controller, take system efficiency, economic benefits and environmental impact as optimization goals, and output the optimized steam extraction amount, energy storage amount and heat release amount control strategy;

[0035] Based on historical operation data, the pre-trained steam load prediction model is used to calculate the predicted steam load. At the same time, the actual steam load data of the system is collected in real time through the sensor network. The predicted load output by the steam load prediction model is compared with the actual steam load data collected in real time to calculate the prediction deviation.

[0036] The calculated prediction deviation is input into the model predictive controller, and combined with the current state of the system, the steam extraction amount, energy storage and heat release are re-optimized to generate an updated control strategy;

[0037] Utilizing the real-time collected operating data and calculated prediction deviation, the parameters of the steam load prediction model are updated through an online learning algorithm. At the same time, the high-precision dynamic model parameters of the extraction steam energy storage and heat release system are updated. Based on the updated system dynamic model and the updated control strategy, a real-time optimization algorithm is used to solve the model predictive control problem, generate real-time control instructions for the extraction steam volume, energy storage and heat release, execute the generated real-time control instructions, adjust the system's extraction steam volume, energy storage and heat release, and realize closed-loop feedback control of the extraction steam energy storage and heat release system.

[0038] According to a second aspect of the embodiments of the present invention,

[0039] Provided is a steam extraction energy storage and heat release dynamic regulation system in a thermal system, comprising:

[0040] The first unit is used to obtain multi-dimensional operating parameters of the thermal system and heat demand data of the user end, wherein the heat demand data includes user type, historical heat demand curve and real-time heat demand, and analyze the multi-dimensional operating parameters and heat demand data based on the deep learning algorithm to build a thermal system load prediction model. The thermal system load prediction model is used to predict the steam load state in a specific period of time in the future in combination with weather forecast data and user behavior patterns, and the load period is dynamically divided into peak period, flat peak period and valley period according to the predicted steam load state;

[0041] The second unit is used to establish a multi-energy collaborative optimization model based on the divided load periods, obtain the optimal operation strategy for each period by solving the multi-energy collaborative optimization model, and perform steam extraction storage and heat release operations in the corresponding load period according to the optimal operation strategy for each period;

[0042] The third unit is used to establish a closed-loop feedback mechanism during the steam extraction storage and heat release process, compare the deviation between the predicted steam load and the actual load in real time, and dynamically fine-tune the steam extraction, storage and heat release. At the same time, the load prediction model is continuously optimized according to the deviation data. Based on the optimized load prediction model, the real-time operating status of the thermal system, the actual heat usage at the user end and the energy market price information are integrated to construct a multi-objective optimization model. The multi-objective optimization model is used to adaptively adjust the operating parameters of steam extraction storage and heat release to achieve dynamic regulation of steam extraction storage and heat release in the thermal system.

[0043] According to a third aspect of the embodiments of the present invention,

[0044] An electronic device is provided, comprising:

[0045] processor;

[0046] a memory for storing processor-executable instructions;

[0047] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0048] According to a fourth aspect of the embodiments of the present invention,

[0049] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0050] In this embodiment, load forecasting and multi-energy collaborative optimization are used to achieve optimal scheduling and energy utilization of the system, thereby reducing energy waste and equipment loss. Through dynamic load division and closed-loop feedback mechanism, the operation strategy is adaptively adjusted to cope with load fluctuations and uncertainties, thereby ensuring stable operation of the system. Through a multi-objective optimization model, energy prices and system efficiency are comprehensively considered, the operating parameters of steam extraction storage and heat release are optimized, and energy procurement and equipment maintenance costs are minimized. By optimizing scheduling and energy configuration, the consumption of fossil fuels is reduced, carbon emissions and pollutant emissions are reduced, and clean and efficient operation of the thermal system is achieved. Through accurate load forecasting and dynamic regulation, users are provided with stable and reliable heat supply, improving user satisfaction and comfort. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a schematic flow chart of a method for dynamically regulating steam extraction energy storage and heat release in a thermal system according to an embodiment of the present invention;

[0052] Figure 2 It is a structural schematic diagram of the steam extraction energy storage and heat release dynamic regulation system in the thermal system of an embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0055] Figure 1 FIG. 1 is a flow chart of a method for dynamically regulating steam extraction energy storage and heat release in a thermal system according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0056] S101. Obtain multi-dimensional operating parameters of the thermal system and heat demand data on the user side, wherein the heat demand data includes user type, historical heat consumption curve and real-time heat consumption. Based on a deep learning algorithm, analyze the multi-dimensional operating parameters and heat demand data to construct a thermal system load prediction model. Utilize the thermal system load prediction model, combine weather forecast data and user behavior patterns, and predict the steam load state within a specific period in the future. According to the predicted steam load state, dynamically divide the load time periods into peak period, off-peak period and valley period.

[0057] Among them, the multi-dimensional operating parameters include the actual load of the thermal system, steam pressure, steam temperature, steam flow and other key indicators that reflect the operating status of the system. The heat demand data includes user type, historical heat consumption curve and real-time heat consumption, which comprehensively reflect the characteristics of heat demand on the user side. After obtaining the above data, the deep learning algorithm is used to comprehensively analyze the multi-dimensional operating parameters and heat demand data. Specifically, the long short-term memory network (LSTM) is used to extract and learn the features of time series data to explore the intrinsic relationship between load changes and influencing factors. At the same time, the convolutional neural network (CNN) is used to identify the user's heat behavior pattern and extract high-level semantic features such as user heat habits and periodicity. Based on LSTM and CNN, a multi-task learning framework is constructed to jointly train the two subtasks of load prediction and user behavior recognition to achieve feature sharing and information interaction, which significantly improves the accuracy and robustness of load prediction.

[0058] Through the deep learning algorithm, a load forecasting model for the thermal system was obtained. The model comprehensively considers the operating characteristics of the system itself and the heat consumption behavior on the user side, and can accurately characterize the load change law. When applying this model, this method also introduces weather forecast data as an important external influencing factor. By analyzing the correlation between weather conditions and load, taking meteorological elements such as temperature, humidity, and wind speed as inputs of the model, the load forecast results are dynamically adjusted, and the forecast accuracy is further improved. In addition, considering the uncertainty of user behavior, this method combines user profiling technology to characterize the user's typical heat consumption pattern based on user attributes and historical heat consumption data, and introduces it as prior knowledge into the load forecasting model, which enhances the model's adaptability to personalized heat consumption behavior.

[0059] Based on the load prediction model of the thermal system, the steam load status in a specific period of time in the future can be predicted. In order to facilitate the formulation of the steam extraction energy storage strategy, this method adaptively divides the load into peak period, flat peak period and valley period according to the prediction results. The division method takes into account the dynamic change characteristics of the load, and adopts the sliding window technology to continuously analyze the load level in the future period in units of hours. According to the relative high and low load and the change trend, the division boundary of each time period is dynamically adjusted. At the same time, the fuzzy clustering algorithm is introduced to map the load status into three fuzzy subsets of peak, flat peak and valley, and the membership degree of each time period is obtained, which provides flexibility and adaptability for the subsequent formulation of steam extraction energy storage strategy.

[0060] In an optional embodiment,

[0061] Based on deep learning algorithms, we analyze multi-dimensional operating parameters and heat demand data to build a thermal system load prediction model. We use the thermal system load prediction model, combined with weather forecast data and user behavior patterns, to predict the steam load state in a specific period of time in the future. Based on the predicted steam load state, we dynamically divide the load period into peak period, off-peak period and valley period.

[0062] Preprocess the multi-dimensional operating parameters of the thermal system and the heat demand data of the user end to generate a standardized multi-source data set, extract knowledge elements from the standardized multi-source data set using natural language processing technology, and construct a multi-level knowledge graph including the equipment layer, system layer and application layer;

[0063] Based on the multi-level knowledge graph, the thermal system is abstracted into a dynamic weighted graph network, in which nodes represent system components and edges represent physical connections and energy flows between components. A hierarchical recurrent neural network is constructed for the multi-scale time series data in the standardized multi-source data set, and the semantic information in the multi-level knowledge graph is used as additional input to extract multi-scale time series features.

[0064] Design a multimodal deep fusion network to adaptively fuse system topology features, timing features, and semantic features to build a load forecasting model. Use a multi-scale convolutional neural network to extract features from user heat data. Combine user attribute information in a multi-level knowledge graph, and dynamically fuse the extracted features through an attention mechanism to build a user behavior model.

[0065] The load forecasting basic model, user behavior model and anomaly detection model are integrated into a multi-task learning framework, and combined with the semantic enhancement features to generate multi-time scale load forecasting sequences. The load forecasting sequences are divided into load periods, including peak period, off-peak period and valley period, using an adaptive load partitioning algorithm.

[0066] For example, first, the multi-source heterogeneous data of the thermal system, including equipment operating parameters, heat meter readings, meteorological data, user heat records, etc. collected by the SCADA system, are pre-processed by data cleaning and normalization to generate a standardized multi-source data set. Then, natural language processing techniques, such as named entity recognition and relationship extraction, are used to automatically extract knowledge elements such as system components, attributes, and relationships from structured and unstructured data, and a multi-level knowledge graph of the device layer, system layer, and application layer is constructed through entity linking and relationship reasoning to form an abstract representation of the global semantic information of the thermal system.

[0067] On this basis, the thermal system is abstracted into a dynamic weighted graph network model. The nodes in the graph represent the equipment, pipelines and other components in the system, and the edges represent the physical connection relationship and thermodynamic parameters between the components. The attributes of the nodes come from the static attributes and dynamic operation parameters of the equipment in the knowledge graph, and the weights of the edges reflect the energy transfer efficiency and impact intensity between components. For the graph network structure data, a multi-layer graph convolutional network is designed to aggregate and propagate node information by sliding the convolution kernel on the graph domain, capturing the topological structure characteristics of the system. At the same time, the graph attention mechanism is introduced to adaptively assign weights to different nodes, highlight key nodes and related paths, and improve the efficiency and accuracy of information transmission.

[0068] While modeling the graph network, a hierarchical recurrent neural network is constructed for multi-scale time series data in multi-source data sets, such as hourly, daily, and weekly load data. The bottom layer uses the long short-term memory network LSTM to characterize short-term time series features and intraday fluctuation patterns, and the upper layer uses the gated recurrent unit GRU to model long-term trend changes and periodic patterns. At each time step, the multi-source time series data at the current moment is combined with the semantic information in the knowledge graph, and the entity embedding vector is introduced as an additional input to enhance the expression ability of time series features.

[0069] After obtaining the system topology characteristics and multi-scale time series characteristics, a multi-modal deep fusion network is designed to achieve dynamic interaction and fusion of different modal features through a multi-head attention mechanism. Specifically, the characteristics of each modality are first sent to an independent feedforward neural network for transformation, and then the correlation between different modalities is calculated through the attention matrix to generate fusion weights, and finally the weighted sum is obtained to obtain a unified multi-modal feature representation. On this basis, the fully connected layer is connected to build a basic model for load forecasting, realizing end-to-end modeling and optimization.

[0070] In order to further improve the prediction accuracy and generalization performance, this paper also introduces user behavior modeling and anomaly detection modules. For user heat usage data, a multi-scale one-dimensional convolutional neural network is used to extract user heat usage patterns on different time scales. Then, combined with the user attribute information in the knowledge graph, the influence weights of different users are dynamically adjusted through the attention mechanism to characterize the group heat usage behavior characteristics. At the same time, based on unsupervised anomaly detection algorithms such as isolation forests, anomalies in load data are identified and filtered to improve data quality. Finally, the three tasks of load forecasting, user behavior modeling, and anomaly detection are integrated into a unified multi-task learning framework. By sharing hidden layer parameters, knowledge transfer and feature enhancement between different tasks are achieved, which improves the learning efficiency and generalization ability of the model.

[0071] In the application stage, external factors such as weather forecast data and holiday information are combined with the environment-load association rules in the knowledge graph to generate semantically enhanced features. The feature weights are dynamically adjusted through the attention mechanism and the gating mechanism, and introduced into the optimized load forecasting model to generate a multi-time scale load forecast sequence for a period of time in the future. On this basis, combined with historical load data and typical load patterns summarized in the knowledge graph, an adaptive load partitioning algorithm is used to divide the load sequence into basic load, peak load, valley load, etc. Fuzzy C-means clustering is used to cluster different types of loads, and probabilistic graph models such as Gaussian mixture models are constructed to estimate the probability density functions of various types of loads and generate load distribution prediction results.

[0072] In this embodiment, by integrating knowledge graphs and deep learning technology, multi-level semantic information is used to guide the construction of load forecasting models, overcoming the limitations of traditional data-driven methods and giving the model certain physical meaning and interpretability. A variety of deep learning models such as graph networks, time series networks, and fusion networks are designed specifically to extract load influencing factors from three dimensions: space, time, and multimodality, which can fully explore the multi-scale and nonlinear patterns contained in load data. Load forecasting, user modeling, and anomaly detection are integrated into a multi-task learning paradigm to enhance the synergy and complementarity between different tasks, simplifying the modeling process while improving generalization performance. By analyzing the predicted data, the model can automatically identify and divide different load periods to support subsequent scheduling and optimization decisions.

[0073] In an optional embodiment,

[0074] The load forecast sequence is divided into the following steps using an adaptive load division algorithm, and the load period division results generated include:

[0075] The load forecasting sequence is subjected to wavelet transformation to obtain multi-layer scale coefficients and fluctuation coefficients. Adaptive weights are introduced to perform weighted fusion on the obtained multi-layer scale coefficients and fluctuation coefficients to obtain a feature sequence. The obtained feature sequence is used as a sample input into a pre-built process mixture model for clustering to obtain clustering results and cluster centers.

[0076] The typical load pattern with the highest similarity to the cluster center is retrieved from the pre-constructed multi-level knowledge graph, and the similarity between the cluster center and the predefined load pattern is calculated by cosine similarity. For each cluster, the predefined load pattern with the highest similarity is selected as the semantic label. Based on the obtained clustering results and semantic labels, a time period division result is generated for each moment in the load sequence, and according to the differences in load characteristics in different time periods, the generated time period division result is fine-tuned to obtain the final load time period division result.

[0077] For example, the load forecast sequence is first subjected to wavelet transform, and the load sequence is decomposed at multiple scales to obtain multi-layer scale coefficients and fluctuation coefficients reflecting different frequency characteristics. In order to make full use of the information contained in each layer of scale and fluctuation coefficient, adaptive weights are introduced to perform weighted fusion of multi-layer coefficients. The adaptive weights can be dynamically adjusted according to the characterization ability of each layer of coefficients on load characteristics. Through weighted fusion, a comprehensive feature sequence is obtained, which fully integrates multi-scale information.

[0078] The obtained characteristic sequence is used as a sample input into the pre-built process mixture model for clustering. The process mixture model is a flexible probabilistic model that can fit complex data distributions by mixing multiple Gaussian processes. By clustering the characteristic sequence with the process mixture model, the load sequence can be divided into multiple clusters with similar characteristics, and the clustering results and the center of each cluster can be obtained.

[0079] In order to give the clustering results a clear physical meaning, the typical load pattern with the highest similarity to the cluster center is retrieved from the pre-built multi-level knowledge graph. The multi-level knowledge graph is a structured knowledge representation method that describes the hierarchical semantic information of the load pattern through multi-level concept nodes and relationship edges. The similarity between the cluster center and the predefined load pattern is calculated by cosine similarity. For each cluster, the predefined load pattern with the highest similarity is selected as the semantic label to give the clustering results a clear physical interpretation.

[0080] Based on the obtained clustering results and semantic labels, a preliminary time period division result is generated for each moment in the load sequence. The load sequence is divided into multiple time periods by clustering. Each time period corresponds to a cluster cluster and semantic label, which reflects the load characteristics and patterns of the time period. Considering that the load characteristics of different time periods may be different, the preliminary generated time period division results are fine-tuned. By analyzing the load characteristics of adjacent time periods, such as the average load level and fluctuation degree, the time period boundaries are appropriately adjusted so that the divided time periods more accurately reflect the load change law and obtain the final load time period division results.

[0081] Assume that the load forecast sequence of an industrial park is divided into time periods. First, the load sequence is subjected to wavelet transform to obtain multi-layer scale coefficients and fluctuation coefficients, and the feature sequence is obtained by adaptive weighted fusion. Then the feature sequence is input into the pre-trained process mixture model for clustering to obtain several clusters and cluster centers. The typical load patterns with the highest similarity to the cluster center are retrieved from the pre-built industrial park load pattern knowledge graph, such as "peak production period" and "equipment maintenance period", as the semantic labels of the clusters. Based on the clustering results and semantic labels, the load sequence is divided into multiple time periods, each corresponding to a load pattern. Finally, according to the differences in load characteristics between adjacent time periods, such as the obvious differences in load levels and fluctuations between the peak production period and the equipment maintenance period, the time period boundaries are fine-tuned to obtain the final load time period division results.

[0082] In this embodiment, through wavelet transform and adaptive weight fusion, multi-scale information is fully utilized to extract the comprehensive characteristics of the load sequence, providing an effective sample representation for subsequent clustering. By clustering the feature sequence using a process mixture model, the potential pattern in the load sequence can be discovered, and the load sequence can be divided into multiple time periods with similar characteristics to achieve load time period division. Through multi-level knowledge graphs and similarity calculations, the clustering results are given a clear physical meaning, making the time period division results interpretable and convenient for engineering applications. The initially generated time period division results are fine-tuned to consider the differences in load characteristics in different time periods to improve the accuracy and granularity of the time period division. The load sequence is automatically divided into time periods, which reduces the workload of manual analysis, improves efficiency and reliability, and provides a basis for subsequent load analysis, prediction, and optimization.

[0083] S102. Based on the divided load periods, a multi-energy collaborative optimization model is established. By solving the multi-energy collaborative optimization model, the optimal operation strategy for each period is obtained. According to the optimal operation strategy for each period, steam extraction storage and heat release operations are performed in the corresponding load period.

[0084] Among them, the multi-energy collaborative optimization model comprehensively considers the operating characteristics of the thermal system, the heat demand on the user side and the price information of the energy market, and obtains the optimal operation strategy for each time period through mathematical modeling and optimization solution. The optimal operation strategy includes the optimal configuration of steam extraction, energy storage and heat release, as well as the corresponding equipment operation parameters and scheduling schemes. According to the optimal operation strategy for each time period, steam extraction, energy storage and heat release operations are performed in the corresponding load period. By extracting steam for energy storage during peak periods and releasing heat for supply during valley periods, load peak shaving and valley filling can be achieved, thereby improving the operating efficiency and economy of the system.

[0085] In an optional embodiment,

[0086] Based on the divided load periods, a multi-energy collaborative optimization model is established. By solving the multi-energy collaborative optimization model, the optimal operation strategy for each period is obtained. According to the optimal operation strategy for each period, steam extraction storage and heat release operations are performed in the corresponding load period, including:

[0087] According to the peak load period obtained, the steam supply gap at each moment is predicted, and a peak heat release optimization model is established with the goal of minimizing the user-side heat supply gap. The peak heat release optimization model is solved using the model predictive control algorithm to obtain the optimal heat release power at each moment, generate peak heat release instructions, control the energy storage device to release stored energy, and supplement the peak steam supply gap;

[0088] Monitor the load fluctuation during the divided off-peak load period. When the actual load deviates from the predicted load by more than the set threshold, the steam extraction or heat release regulation mechanism is triggered, and the control algorithm is used to calculate the regulation amount to smooth the load fluctuation.

[0089] According to the divided off-peak load period, the steam extraction amount at each moment is predicted, and an optimization model for off-peak steam extraction with the goal of maximizing the energy storage is established. The dynamic programming algorithm is used to solve the off-peak steam extraction optimization model, and the optimal steam extraction power at each moment is obtained. The off-peak steam extraction command is generated, and the steam extraction device is controlled to store the remaining steam in the energy storage device.

[0090] The price signals of electricity and gas market are introduced to build a multi-energy collaborative optimization model. The mixed integer nonlinear programming algorithm is used to solve the multi-energy collaborative optimization model to obtain the optimal day-ahead dispatch plan for electricity, gas and heat. Based on the optimal day-ahead dispatch plan obtained, the real-time electricity price and gas price information is used, and a rolling time domain optimization algorithm is used to update the load forecast and market price in each dispatching period to obtain the optimal operation strategy for each period. According to the optimal operation strategy for each period, steam extraction storage and heat release operations are performed in the corresponding load period.

[0091] For example, for the peak load period obtained by division, the steam supply gap at each moment is first predicted. The steam supply gap refers to the difference between the steam supply of the thermal system and the actual heat demand of the user, reflecting the degree of imbalance between supply and demand. By analyzing the historical load data and the user's heat consumption mode, the load prediction model is used to predict the heat demand at each moment of the peak period, and compared with the steam supply capacity of the system to obtain the predicted value of the steam supply gap. A peak heat release optimization model with the goal of minimizing the heat supply gap on the user side is established. The model takes minimizing the heat supply gap as the optimization goal, and satisfies the user's heat demand to the maximum extent by reasonably scheduling the heat release power of the energy storage device. The peak heat release optimization model is solved by the model predictive control algorithm to obtain the optimal heat release power at each moment. Model predictive control is a model-based optimization control method that optimizes the control strategy for a period of time in the future in a rolling manner and continuously updates the optimization results according to the actual measurement values. It is predictive and adaptive. According to the optimization results, a peak heat release instruction is generated to control the energy storage device to release the stored energy, supplement the peak steam supply gap, and ensure the heating quality on the user side.

[0092] Focus on load fluctuations during the divided off-peak load period. Determine the deviation between the actual load and the predicted load by real-time monitoring of the system's operating parameters and the heat consumption on the user side. When the deviation exceeds the preset threshold, the steam extraction or heat release regulation mechanism is triggered. Use control algorithms, such as PID control or fuzzy control, to calculate the required regulation amount, smooth the load fluctuations by adjusting the steam extraction or heat release, and maintain stable operation of the system.

[0093] For the divided off-peak load period, the focus is on predicting the amount of steam that can be extracted at each moment. The amount of steam that can be extracted refers to the amount of steam that can be extracted from the steam network by the thermal system for energy storage or other purposes under the premise of ensuring the heat demand of users. The amount of steam that can be extracted at each moment is predicted by analyzing the load characteristics and equipment operation status of the off-peak period. A off-peak steam extraction optimization model with the goal of maximizing the energy storage is established. The model takes extracting as much steam as possible for energy storage during the off-peak period as the optimization goal, while considering factors such as equipment capacity and safety constraints. The off-peak steam extraction optimization model is solved by the dynamic programming algorithm to obtain the optimal steam extraction power at each moment. Dynamic programming is an optimization algorithm based on optimal substructure and overlapping subproblems. It decomposes the problem into interrelated subproblems and uses the optimal solution of the subproblem to construct the optimal solution of the original problem. According to the optimization results, the off-peak steam extraction instruction is generated to control the steam extraction device to store the remaining steam in the energy storage device to provide energy support for the heat release during the peak period.

[0094] In order to further improve the economy and flexibility of the system, the price signals of electricity and gas market are introduced to construct a multi-energy collaborative optimization model. The multi-energy collaborative optimization model comprehensively considers multiple energy forms such as electricity, gas, and heat, as well as the mutual influence and constraint relationship between them, and optimizes the scheduling and configuration of each energy source to achieve efficient energy utilization and cost minimization. The mixed integer nonlinear programming algorithm is used to solve the multi-energy collaborative optimization model to obtain the optimal day-ahead scheduling plan for electricity, gas, and heat. Mixed integer nonlinear programming is a mathematical optimization method that can handle complex optimization problems containing continuous variables, discrete variables, and nonlinear constraints. Based on the optimal day-ahead scheduling plan obtained, the real-time electricity price and gas price information are used, and the rolling time domain optimization algorithm is used to update the load forecast and market price in each scheduling period to obtain the optimal operation strategy for each period. Rolling time domain optimization is a dynamic optimization method that continuously optimizes the scheduling strategy for a period of time in the future, and updates the optimization results according to real-time information to adapt to real-time changes in load and price. According to the optimal operation strategy of each period, steam extraction and heat release operations are performed in the corresponding load period to achieve dynamic optimization and regulation of the thermal system.

[0095] In this embodiment, by optimizing and controlling in different time periods, corresponding optimization strategies are adopted according to the characteristics of different load periods to maximize the use of energy, reduce waste, and reduce operating costs. By real-time monitoring of load fluctuations, the steam extraction or heat release adjustment mechanism is triggered to dynamically respond to load changes and maintain the stable operation of the system. At the same time, a rolling time domain optimization algorithm is adopted to update the scheduling strategy in real time to adapt to real-time changes in load and price. Through peak heat release optimization, the steam supply gap is predicted, the energy storage device is reasonably scheduled, the peak heat demand is supplemented, the user's heat demand is met to the maximum extent, and user satisfaction is improved. Through the multi-energy collaborative optimization model, various energy forms such as electricity, gas, and heat are comprehensively considered to optimize the scheduling and configuration of energy, give play to the complementary advantages between energy sources, and improve energy utilization efficiency. Through the optimization of steam extraction during the trough period, the remaining steam is stored to provide energy support for heat release during the peak period, and the cascade utilization of energy is realized. At the same time, energy configuration is optimized, fossil fuel consumption is reduced, carbon emissions are reduced, and the goal of energy conservation and emission reduction is achieved.

[0096] In an optional embodiment,

[0097] Introduce electricity and gas market price signals, build a multi-energy collaborative optimization model, use mixed integer nonlinear programming algorithm to solve the multi-energy collaborative optimization model, and obtain the optimal day-ahead dispatch plan for electricity, gas, and heat. Based on the optimal day-ahead dispatch plan, use real-time electricity and gas price information, and use a rolling time domain optimization algorithm to update load forecasts and market prices in each dispatch period to obtain the optimal operation strategy for each period. According to the optimal operation strategy for each period, steam extraction storage and heat release operations are performed in the corresponding load period, including:

[0098] Construct a multi-energy collaborative optimization model, which takes into account the coupling characteristics of the thermal system, the power system and the gas system, and uses the concept of energy hub to abstract the three systems into a multi-energy flow network to describe the conversion and transmission relationship between various energy forms;

[0099] Introducing price signals from the electricity and gas markets, building a multi-objective optimization model, the objective functions of which include minimizing operating costs, maximizing energy efficiency, and optimizing environmental benefits;

[0100] Establish constraints for multi-energy collaborative optimization, including power system constraints, gas system constraints, thermal system constraints, energy hub constraints, and electric-thermal coupling constraints;

[0101] A mixed integer nonlinear programming algorithm is used to solve the multi-energy collaborative optimization model, wherein the problem is decomposed into a main problem and sub-problems using a Benders decomposition algorithm, and an optimal day-ahead dispatch plan for the power system, the gas system, and the thermal system is generated through iterative solution;

[0102] At the beginning of each scheduling period, the latest electricity price and gas price information is obtained, and the load forecast results are updated. Taking the current time as the starting point and the predetermined time window as the optimization range, a rolling optimization model is constructed, and the executed scheduling decisions are incorporated into the rolling optimization model as known conditions. The rolling optimization model is solved using a rolling time domain optimization algorithm to obtain the optimal scheduling strategy in the future time window, including the output adjustment and energy flow distribution of the power system, gas system and thermal system. According to the optimal scheduling strategy, steam extraction storage and heat release operations are performed in the corresponding load period.

[0103] Exemplarily, first, a multi-energy collaborative optimization model is constructed, taking into account the coupling characteristics of thermal systems, power systems, and gas systems. The concept of energy hubs is adopted to abstract the three systems into a multi-energy flow network. Energy hubs refer to nodes that can realize conversion and transmission between different energy forms, such as cogeneration units, gas turbines, electric boilers, etc. Through energy hubs, the conversion relationship between energy forms such as electricity, heat, and gas can be described, such as gas turbines converting gas into electricity and heat, and electric boilers converting electricity into heat. At the same time, the losses and constraints of each energy form in the transmission process are considered, such as the line capacity limit of power transmission, the pressure constraint of the gas pipeline network, etc., to construct a complete multi-energy flow network model.

[0104] Secondly, the price signals of the electricity market and the gas market are introduced to construct a multi-objective optimization model. The objective functions include minimizing operating costs, maximizing energy efficiency, and optimizing environmental benefits. Minimizing operating costs means minimizing the total operating costs of the system, including fuel costs, operation and maintenance costs, and start-up and shutdown costs, by optimizing the scheduling and configuration of energy while meeting the energy needs of users. Maximizing energy efficiency means making full use of the complementarity and synergy between energy sources, improving the comprehensive utilization efficiency of energy, and reducing energy waste. Optimizing environmental benefits means minimizing the carbon emissions and pollutant emissions of the energy system while meeting environmental constraints to achieve green and low-carbon operation.

[0105] Then, the constraints for multi-energy collaborative optimization are established, including power system constraints, gas system constraints, thermal system constraints, energy hub constraints and electric-thermal coupling constraints. Power system constraints include power balance constraints, spare capacity constraints, line flow constraints, etc., to ensure the safe and stable operation of the power system. Gas system constraints include gas balance constraints, pressure constraints, pipeline flow constraints, etc., to ensure the normal supply of the gas system and pipeline safety. Thermal system constraints include thermal balance constraints, pipeline flow constraints, supply and return water temperature constraints, etc., to ensure the stable operation of the thermal system and the heating quality on the user side. Energy hub constraints describe the conversion relationship and efficiency limitations between different energy forms, such as the output characteristics of cogeneration units, the start and stop constraints of gas turbines, etc. Electric-thermal coupling constraints characterize the coupling relationship between the power system and the thermal system, such as the relationship between the extraction pressure and heating supply of cogeneration units.

[0106] Next, a mixed integer nonlinear programming algorithm is used to solve the multi-energy collaborative optimization model. Since the model contains continuous variables, discrete variables and nonlinear constraints, it is a mixed integer nonlinear programming problem and is difficult to solve. In order to improve the solution efficiency, the Benders decomposition algorithm is used to decompose the problem into a main problem and sub-problems. The main problem is responsible for the overall scheduling decision of energy, such as unit combination, energy flow distribution, etc., while the sub-problem is responsible for the detailed operation optimization of each subsystem, such as power flow calculation of the power system, pipeline simulation of the gas system, etc. By iteratively solving the main problem and sub-problems, and continuously updating the cutting plane of the main problem, the optimal day-ahead scheduling plan for the power system, gas system and thermal system is finally obtained, including the output scheduling of each unit, the distribution plan of the energy flow, etc.

[0107] Finally, in the actual operation process, the rolling time domain optimization method is used to dynamically adjust the optimization strategy. At the beginning of each scheduling period, the latest electricity price and gas price information are obtained, and the load forecast results are updated at the same time. Taking the current time as the starting point and the predetermined time window (such as 24 hours) as the optimization range, a rolling optimization model is constructed. The executed scheduling decisions are included in the rolling optimization model as known conditions, and the current price signals and load forecasts are considered at the same time. The rolling optimization model is solved using a rolling time domain optimization algorithm. The optimal scheduling strategy in the future time window is obtained, including the output adjustment and energy flow distribution scheme of the power system, gas system and thermal system. According to the optimal scheduling strategy, steam extraction and heat release operations are performed in the corresponding load period to achieve multi-energy collaborative optimization scheduling.

[0108] In this embodiment, through multi-energy collaborative optimization, the complementarity and synergy between different energy sources are fully utilized, the scheduling and configuration of energy are optimized, energy waste is reduced, and energy utilization efficiency is improved. By introducing price signals from the electricity and gas markets, the economic scheduling of energy is optimized, and the total operating cost of the system is minimized while meeting user needs, and energy cost expenditure is reduced. The rolling time domain optimization method is adopted to dynamically adjust the optimization strategy according to real-time price and load forecasts to adapt to fluctuations in energy prices and user needs, thereby improving the flexibility and adaptability of the system. Through optimized scheduling, clean energy such as gas and electricity are maximized, the consumption of fossil fuels is reduced, carbon emissions and pollutant emissions are reduced, and green and low-carbon operation is achieved. By considering the operating constraints and safety restrictions of the power system, gas system, and thermal system, the stable operation of each energy system is ensured, and the reliability and safety of energy supply are improved. Through the multi-energy collaborative optimization model, the impact of different energy equipment combinations and operating modes on system efficiency, cost, environment, etc. can be analyzed, providing a quantitative decision-making basis for the planning and design of the energy system.

[0109] S103. During the steam extraction and heat release process, a closed-loop feedback mechanism is established to compare the deviation between the predicted steam load and the actual load in real time, and dynamically fine-tune the steam extraction, storage and heat release. At the same time, the load prediction model is continuously optimized according to the deviation data. Based on the optimized load prediction model, the real-time operating status of the thermal system, the actual heat usage at the user end and the energy market price information are integrated to construct a multi-objective optimization model. The multi-objective optimization model is used to adaptively adjust the operating parameters of steam extraction and heat release to achieve dynamic regulation of steam extraction and heat release in the thermal system.

[0110] First, a real-time monitoring system is established and sensors are deployed to obtain real-time steam load data, including the heat consumption at the user end and the operating parameters of the system. These data are aggregated to the central control platform through the data acquisition system. Next, a load forecasting module is designed to use historical heat consumption data and real-time meteorological information (such as temperature, humidity, etc.) to forecast steam load. This module generates forecast values ​​for future loads through a deep learning algorithm. During operation, the real-time monitoring module regularly compares the predicted load with the actual load and calculates the load deviation. This deviation reflects the difference between the forecast and the actual, providing a basis for subsequent adjustments.

[0111] According to the calculated deviation, dynamic adjustment is implemented. The system fine-tunes the steam extraction, energy storage and heat release in real time according to the preset adjustment strategy. For example, in the case of large load deviation, the system can automatically increase the steam extraction to meet the immediate demand. At the same time, the deviation data will be recorded and fed back to the load forecasting model. This feedback mechanism prompts the model to continuously learn and optimize to improve the accuracy of future forecasts. The model will analyze the rules of the deviation data, adjust its parameters and structure, and gradually reduce the deviation in future forecasts. In this process, the real-time operating status of the thermal system, the heat consumption of the user end and the energy market price information are integrated to construct a multi-objective optimization model. The model will comprehensively consider multiple objectives such as operating costs, energy efficiency and environmental impact, and adaptively adjust the operating parameters of steam extraction, energy storage and heat release through optimization algorithms to achieve the best system performance. Finally, the system will regularly evaluate the optimization results, and further iterate and improve the control strategy through operation analysis and user feedback to ensure that the thermal system can operate efficiently and stably.

[0112] In an optional embodiment,

[0113] During the steam extraction and heat release process, a closed-loop feedback mechanism is established to compare the deviation between the predicted steam load and the actual load in real time, and dynamically adjust the steam extraction, energy storage and heat release, including:

[0114] Establish a high-precision dynamic model of the steam extraction energy storage and heat release system, which takes the steam extraction amount, energy storage amount and heat release amount as input variables, and takes system efficiency, energy consumption and steam output as output variables. Based on the established high-precision dynamic model, design a model predictive controller, take system efficiency, economic benefits and environmental impact as optimization goals, and output the optimized steam extraction amount, energy storage amount and heat release amount control strategy;

[0115] Based on historical operation data, the pre-trained steam load prediction model is used to calculate the predicted steam load. At the same time, the actual steam load data of the system is collected in real time through the sensor network. The predicted load output by the steam load prediction model is compared with the actual steam load data collected in real time to calculate the prediction deviation.

[0116] The calculated prediction deviation is input into the model predictive controller, and combined with the current state of the system, the steam extraction amount, energy storage and heat release are re-optimized to generate an updated control strategy;

[0117] Utilizing the real-time collected operating data and calculated prediction deviation, the parameters of the steam load prediction model are updated through an online learning algorithm. At the same time, the high-precision dynamic model parameters of the extraction steam energy storage and heat release system are updated. Based on the updated system dynamic model and the updated control strategy, a real-time optimization algorithm is used to solve the model predictive control problem, generate real-time control instructions for the extraction steam volume, energy storage and heat release, execute the generated real-time control instructions, adjust the system's extraction steam volume, energy storage and heat release, and realize closed-loop feedback control of the extraction steam energy storage and heat release system.

[0118] For example, first of all, it is key to establish a high-precision dynamic model. This model needs to be designed in detail according to the operating characteristics of the system, and the input variables include steam extraction (referring to the amount of steam extracted from the boiler or steam system), energy storage (the amount used to store energy), and heat release (the amount of heat released to the environment). These input variables affect the overall efficiency and energy consumption of the system, thereby affecting the steam output.

[0119] Secondly, based on the established dynamic model, a model predictive controller is designed. This controller takes system efficiency, economic benefits and environmental impact as optimization objectives, and comprehensively considers the best operation strategy under different operating conditions. System efficiency in the optimization objective refers to the input energy required for unit energy output, while economic benefits are usually related to operating costs and benefits, and environmental impact reflects the degree of negative impact on the environment, such as emissions and resource consumption.

[0120] Next, we use historical operation data to predict steam load. Through the pre-trained steam load prediction model, we can calculate the predicted steam load in the future. In this process, the sensor network is used to collect the actual steam load data of the system in real time, and the predicted load is compared with the actual load to calculate the deviation between the two. This prediction deviation provides a basis for subsequent control strategy adjustments.

[0121] Based on the calculated prediction deviation, the model predictive controller is input to re-optimize the steam extraction, energy storage and heat release with the current system status as reference. In this way, we can generate an updated control strategy to ensure that the system operates in the best state.

[0122] Using real-time collected data and calculated prediction deviations, online learning algorithms are used to continuously update the parameters of the steam load prediction model. This process ensures the adaptability of the model, enabling it to cope with dynamically changing system environments. At the same time, the updated high-precision dynamic model parameters will be combined with the control strategy, and the real-time optimization algorithm will be used to solve the model predictive control problem, and ultimately generate specific real-time control instructions for steam extraction, energy storage, and heat release.

[0123] Finally, the generated real-time control instructions are executed to adjust the system parameters and realize closed-loop feedback control of the steam extraction storage and heat release system. Through this process, we can not only optimize the operating efficiency of the system, but also improve economic benefits, reduce environmental impact, and form a virtuous cycle management model.

[0124] In this embodiment, through real-time dynamic adjustment, the system instability caused by load fluctuations is reduced and the smoothness of operation is improved. Through the closed-loop feedback mechanism and continuous learning of historical data, the accuracy of the load forecasting model is enhanced and the forecast error is reduced. Effective coordination and allocation of different energy forms are achieved to ensure the rational use of various energy resources during high-demand periods to avoid waste of resources. It can quickly respond to market changes and user demand changes, adapt to different operating environments and conditions, and improve the overall flexibility of the system. By optimizing the control strategy, the use of renewable energy, such as solar or wind energy, is promoted to improve the sustainability of the energy system. By integrating advanced data analysis and control technologies, intelligent management of thermal systems is achieved to improve management efficiency and decision-making level.

[0125] Figure 2 FIG. 1 is a schematic diagram of a structure of a steam extraction energy storage and heat release dynamic regulation system in a thermal system according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0126] The first unit is used to obtain multi-dimensional operating parameters of the thermal system and heat demand data of the user end, wherein the heat demand data includes user type, historical heat demand curve and real-time heat demand, and analyze the multi-dimensional operating parameters and heat demand data based on the deep learning algorithm to build a thermal system load prediction model. The thermal system load prediction model is used to predict the steam load state in a specific period of time in the future in combination with weather forecast data and user behavior patterns, and the load period is dynamically divided into peak period, flat peak period and valley period according to the predicted steam load state;

[0127] The second unit is used to establish a multi-energy collaborative optimization model based on the divided load periods, obtain the optimal operation strategy for each period by solving the multi-energy collaborative optimization model, and perform steam extraction storage and heat release operations in the corresponding load period according to the optimal operation strategy for each period;

[0128] The third unit is used to establish a closed-loop feedback mechanism during the steam extraction storage and heat release process, compare the deviation between the predicted steam load and the actual load in real time, and dynamically fine-tune the steam extraction, storage and heat release. At the same time, the load prediction model is continuously optimized according to the deviation data. Based on the optimized load prediction model, the real-time operating status of the thermal system, the actual heat usage at the user end and the energy market price information are integrated to construct a multi-objective optimization model. The multi-objective optimization model is used to adaptively adjust the operating parameters of steam extraction storage and heat release to achieve dynamic regulation of steam extraction storage and heat release in the thermal system.

[0129] According to a third aspect of the embodiments of the present invention,

[0130] An electronic device is provided, comprising:

[0131] processor;

[0132] a memory for storing processor-executable instructions;

[0133] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0134] According to a fourth aspect of the embodiments of the present invention,

[0135] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0136] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0137] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for dynamically regulating steam extraction energy storage and heat release in a thermal system, characterized in that: include: Acquire multi-dimensional operating parameters of the thermal system and heat demand data of the user end, wherein the heat demand data includes user type, historical heat demand curve and real-time heat consumption, analyze the multi-dimensional operating parameters and heat demand data based on the deep learning algorithm, build a thermal system load prediction model, use the thermal system load prediction model, combine weather forecast data and user behavior patterns, predict the steam load state in a specific period of time in the future, and dynamically divide the load period into peak period, flat period and valley period according to the predicted steam load state; Based on the divided load periods, a multi-energy collaborative optimization model is established. By solving the multi-energy collaborative optimization model, the optimal operation strategy for each period is obtained. According to the optimal operation strategy for each period, steam extraction storage and heat release operations are performed in the corresponding load period. During the steam extraction storage and heat release process, a closed-loop feedback mechanism is established to compare the deviation between the predicted steam load and the actual load in real time, and dynamically fine-tune the steam extraction, storage and heat release. At the same time, the load prediction model is continuously optimized according to the deviation data. Based on the optimized load prediction model, the real-time operating status of the thermal system, the actual heat usage at the user end and the energy market price information are integrated to construct a multi-objective optimization model. The multi-objective optimization model is used to adaptively adjust the operating parameters of steam extraction storage and heat release, so as to realize dynamic regulation of steam extraction storage and heat release in the thermal system.

2. The method according to claim 1, characterized in that: Based on deep learning algorithms, we analyze multi-dimensional operating parameters and heat demand data to build a thermal system load prediction model. We use the thermal system load prediction model, combined with weather forecast data and user behavior patterns, to predict the steam load state in a specific period of time in the future. Based on the predicted steam load state, we dynamically divide the load period into peak period, off-peak period and valley period. Preprocess the multi-dimensional operating parameters of the thermal system and the heat demand data of the user end to generate a standardized multi-source data set, extract knowledge elements from the standardized multi-source data set using natural language processing technology, and construct a multi-level knowledge graph including the equipment layer, system layer and application layer; Based on the multi-level knowledge graph, the thermal system is abstracted into a dynamic weighted graph network, in which nodes represent system components and edges represent physical connections and energy flows between components. A hierarchical recurrent neural network is constructed for the multi-scale time series data in the standardized multi-source data set, and the semantic information in the multi-level knowledge graph is used as additional input to extract multi-scale time series features. Design a multimodal deep fusion network to adaptively fuse system topology features, timing features, and semantic features to build a load forecasting model. Use a multi-scale convolutional neural network to extract features from user heat data. Combine user attribute information in a multi-level knowledge graph, and dynamically fuse the extracted features through an attention mechanism to build a user behavior model. The load forecasting basic model, user behavior model and anomaly detection model are integrated into a multi-task learning framework, and combined with the semantic enhancement features to generate multi-time scale load forecasting sequences. The load forecasting sequences are divided into load periods, including peak period, off-peak period and valley period, using an adaptive load partitioning algorithm.

3. The method according to claim 2, characterized in that The load forecast sequence is divided into the following steps using an adaptive load division algorithm, and the load period division results generated include: The load forecasting sequence is subjected to wavelet transformation to obtain multi-layer scale coefficients and fluctuation coefficients. Adaptive weights are introduced to perform weighted fusion on the obtained multi-layer scale coefficients and fluctuation coefficients to obtain a feature sequence. The obtained feature sequence is used as a sample input into a pre-built process mixture model for clustering to obtain clustering results and cluster centers. The typical load pattern with the highest similarity to the cluster center is retrieved from the pre-constructed multi-level knowledge graph, and the similarity between the cluster center and the predefined load pattern is calculated by cosine similarity. For each cluster, the predefined load pattern with the highest similarity is selected as the semantic label. Based on the obtained clustering results and semantic labels, a time period division result is generated for each moment in the load sequence, and according to the differences in load characteristics in different time periods, the generated time period division result is fine-tuned to obtain the final load time period division result.

4. The method according to claim 1, characterized in that: Based on the divided load periods, a multi-energy collaborative optimization model is established. By solving the multi-energy collaborative optimization model, the optimal operation strategy for each period is obtained. According to the optimal operation strategy for each period, steam extraction storage and heat release operations are performed in the corresponding load period, including: According to the peak load period obtained, the steam supply gap at each moment is predicted, and a peak heat release optimization model is established with the goal of minimizing the user-side heat supply gap. The peak heat release optimization model is solved using the model predictive control algorithm to obtain the optimal heat release power at each moment, generate peak heat release instructions, control the energy storage device to release stored energy, and supplement the peak steam supply gap; Monitor the load fluctuation during the divided off-peak load period. When the actual load deviates from the predicted load by more than the set threshold, the steam extraction or heat release regulation mechanism is triggered, and the control algorithm is used to calculate the regulation amount to smooth the load fluctuation. According to the divided off-peak load period, the steam extraction amount at each moment is predicted, and an optimization model for off-peak steam extraction with the goal of maximizing the energy storage is established. The dynamic programming algorithm is used to solve the off-peak steam extraction optimization model, and the optimal steam extraction power at each moment is obtained. The off-peak steam extraction command is generated, and the steam extraction device is controlled to store the remaining steam in the energy storage device. The price signals of electricity and gas market are introduced to build a multi-energy collaborative optimization model. The mixed integer nonlinear programming algorithm is used to solve the multi-energy collaborative optimization model to obtain the optimal day-ahead dispatch plan for electricity, gas and heat. Based on the optimal day-ahead dispatch plan obtained, the real-time electricity price and gas price information is used, and a rolling time domain optimization algorithm is used to update the load forecast and market price in each dispatching period to obtain the optimal operation strategy for each period. According to the optimal operation strategy for each period, steam extraction storage and heat release operations are performed in the corresponding load period.

5. The method according to claim 4, characterized in that Introduce electricity and gas market price signals, build a multi-energy collaborative optimization model, use mixed integer nonlinear programming algorithm to solve the multi-energy collaborative optimization model, and obtain the optimal day-ahead dispatch plan for electricity, gas, and heat. Based on the optimal day-ahead dispatch plan, use real-time electricity and gas price information, and use a rolling time domain optimization algorithm to update load forecasts and market prices in each dispatch period to obtain the optimal operation strategy for each period. According to the optimal operation strategy for each period, steam extraction storage and heat release operations are performed in the corresponding load period, including: Construct a multi-energy collaborative optimization model, which takes into account the coupling characteristics of the thermal system, the power system and the gas system, and uses the concept of energy hub to abstract the three systems into a multi-energy flow network to describe the conversion and transmission relationship between various energy forms; Introducing price signals from the electricity and gas markets, building a multi-objective optimization model, the objective functions of which include minimizing operating costs, maximizing energy efficiency, and optimizing environmental benefits; Establish constraints for multi-energy collaborative optimization, including power system constraints, gas system constraints, thermal system constraints, energy hub constraints, and electric-thermal coupling constraints; A mixed integer nonlinear programming algorithm is used to solve the multi-energy collaborative optimization model, wherein the problem is decomposed into a main problem and sub-problems using a Benders decomposition algorithm, and an optimal day-ahead dispatch plan for the power system, the gas system, and the thermal system is generated through iterative solution; At the beginning of each scheduling period, the latest electricity price and gas price information is obtained, and the load forecast results are updated. Taking the current time as the starting point and the predetermined time window as the optimization range, a rolling optimization model is constructed, and the executed scheduling decisions are incorporated into the rolling optimization model as known conditions. The rolling optimization model is solved using a rolling time domain optimization algorithm to obtain the optimal scheduling strategy in the future time window, including the output adjustment and energy flow distribution of the power system, gas system and thermal system. According to the optimal scheduling strategy, steam extraction storage and heat release operations are performed in the corresponding load period.

6. The method according to claim 1, characterized in that During the steam extraction and heat release process, a closed-loop feedback mechanism is established to compare the deviation between the predicted steam load and the actual load in real time, and dynamically adjust the steam extraction, energy storage and heat release, including: Establish a high-precision dynamic model of the steam extraction energy storage and heat release system, which takes the steam extraction amount, energy storage amount and heat release amount as input variables, and takes system efficiency, energy consumption and steam output as output variables. Based on the established high-precision dynamic model, design a model predictive controller, take system efficiency, economic benefits and environmental impact as optimization goals, and output the optimized steam extraction amount, energy storage amount and heat release amount control strategy; Based on historical operation data, the pre-trained steam load prediction model is used to calculate the predicted steam load. At the same time, the actual steam load data of the system is collected in real time through the sensor network. The predicted load output by the steam load prediction model is compared with the actual steam load data collected in real time to calculate the prediction deviation. The calculated prediction deviation is input into the model predictive controller, and combined with the current state of the system, the steam extraction amount, energy storage and heat release are re-optimized to generate an updated control strategy; Utilizing the real-time collected operating data and calculated prediction deviation, the parameters of the steam load prediction model are updated through an online learning algorithm. At the same time, the high-precision dynamic model parameters of the extraction steam energy storage and heat release system are updated. Based on the updated system dynamic model and the updated control strategy, a real-time optimization algorithm is used to solve the model predictive control problem, generate real-time control instructions for the extraction steam volume, energy storage and heat release, execute the generated real-time control instructions, adjust the system's extraction steam volume, energy storage and heat release, and realize closed-loop feedback control of the extraction steam energy storage and heat release system.

7. A steam extraction storage and heat release dynamic regulation system in a thermal system, used to implement the method described in any one of claims 1 to 6, characterized in that: include: The first unit is used to obtain multi-dimensional operating parameters of the thermal system and heat demand data of the user end, wherein the heat demand data includes user type, historical heat demand curve and real-time heat demand, and analyze the multi-dimensional operating parameters and heat demand data based on the deep learning algorithm to build a thermal system load prediction model. The thermal system load prediction model is used to predict the steam load state in a specific period of time in the future in combination with weather forecast data and user behavior patterns, and the load period is dynamically divided into peak period, flat peak period and valley period according to the predicted steam load state; The second unit is used to establish a multi-energy collaborative optimization model based on the divided load periods, obtain the optimal operation strategy for each period by solving the multi-energy collaborative optimization model, and perform steam extraction storage and heat release operations in the corresponding load period according to the optimal operation strategy for each period; The third unit is used to establish a closed-loop feedback mechanism during the steam extraction storage and heat release process, compare the deviation between the predicted steam load and the actual load in real time, and dynamically fine-tune the steam extraction, storage and heat release. At the same time, the load prediction model is continuously optimized according to the deviation data. Based on the optimized load prediction model, the real-time operating status of the thermal system, the actual heat usage at the user end and the energy market price information are integrated to construct a multi-objective optimization model. The multi-objective optimization model is used to adaptively adjust the operating parameters of steam extraction storage and heat release to achieve dynamic regulation of steam extraction storage and heat release in the thermal system.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images