Machine learning based integrated heat station load prediction method and system

Abstract

The application discloses a machine learning-based integrated heat station load prediction method and system, relates to the technical field of machine learning, and comprises the following steps: acquiring building characteristics and pipeline topological information of a target building to be assembled with an integrated heat station, and collecting a historical meteorological data sequence of an environment where the target building is located; performing first heat station load prediction to obtain a first heat station load, performing integrated heat station basic selection to obtain first selection information; acquiring a group of reference pipeline topological information, performing topological matching degree analysis to obtain a pipeline topological matching degree gradient; obtaining a pipeline maintenance matching degree, combining the pipeline topological matching degree gradient, and calculating a pipeline matching degree to correct the first heat station load and obtain a heat station load. The technical problem that the existing heat station load prediction method is difficult to comprehensively consider complex factors affecting heat load and leads to inaccurate prediction results is solved.

Description

Technical Field

[0001] This application relates to the field of machine learning technology, specifically to an integrated method and system for predicting the load of heating stations based on machine learning. Background Technology

[0002] With the accelerating pace of urbanization and the increasing demands for comfortable living environments, centralized heating systems are playing an increasingly important role in urban infrastructure. Integrated heating stations, as a key component of centralized heating systems, require model selection and design based on load demand. The accuracy of their load forecasting directly impacts the operating efficiency of the heating system, energy consumption, and the user's heating experience.

[0003] However, traditional methods for predicting the load of heating stations often rely on empirical formulas or simple statistical models, which make it difficult to fully consider the complex factors affecting the heating load. This can easily lead to a large deviation between the predicted results and the actual load, resulting in energy waste or a decrease in the quality of heating during the operation of the heating station. Summary of the Invention

[0004] This application provides an integrated heat station load forecasting method and system based on machine learning, which solves the technical problem that existing heat station load forecasting methods cannot fully consider the complex factors affecting heat load, resulting in inaccurate forecasting results.

[0005] The technical solution to the above-mentioned technical problems in this application is as follows:

[0006] In a first aspect, this application provides an integrated heat station load forecasting method based on machine learning, the method comprising:

[0007] Acquire the building characteristics and pipeline topology information of the target building to be assembled with the integrated heating station, and collect the historical meteorological data sequence of the environment in which the target building is located;

[0008] Based on machine learning, the load of the first heating station is predicted according to the building characteristics and historical meteorological data sequence, the load of the first heating station is obtained, and the basic selection of the integrated heating station is carried out to obtain the first selection information.

[0009] Obtain the baseline pipeline topology information group of the integrated heating station, perform topology matching degree analysis with the pipeline topology information, and obtain the pipeline topology matching degree gradient;

[0010] Obtain the pipeline maintenance information of the target building, perform maintenance matching degree analysis based on the pipeline topology matching degree gradient to obtain the pipeline maintenance matching degree, calculate the pipeline matching degree in combination with the pipeline topology matching degree gradient, and correct the load of the first heating station to obtain the heating station load.

[0011] Secondly, this application provides an integrated heat station load forecasting system based on machine learning, including:

[0012] The information acquisition module is used to acquire the building characteristics and pipeline topology information of the target building to be assembled with the integrated heating station, as well as to collect the historical meteorological data sequence of the environment in which the target building is located;

[0013] The model training module is used to predict the load of the first heating station based on machine learning, according to the building characteristics and historical meteorological data sequence, to obtain the load of the first heating station, to perform integrated heating station basic selection, and to obtain the first selection information.

[0014] The data processing module is used to acquire the baseline pipeline topology information group of the integrated heating station, perform topology matching degree analysis with the pipeline topology information, and obtain the pipeline topology matching degree gradient.

[0015] The load correction module is used to acquire the pipeline maintenance information of the target building, perform maintenance matching degree analysis based on the pipeline topology matching degree gradient, obtain the pipeline maintenance matching degree, calculate the pipeline matching degree in combination with the pipeline topology matching degree gradient, and correct the load of the first heating station to obtain the heating station load.

[0016] This application provides one or more technical solutions, which have at least the following technical effects or advantages:

[0017] This application provides an integrated heat station load prediction method and system based on machine learning. First, it acquires the building characteristics and pipeline topology information of the target building, as well as historical meteorological data sequences of the environment in which the target building is located. Second, it uses a pre-built and configured heat load prediction agent to output the first heat station load. This agent is trained on a large amount of sample data to ensure the model's predictive ability. Then, based on the first heat station load, it calls a heat station selection database to perform basic selection, obtaining first selection information and initially determining the range of heat station models. Further, it determines the calling coefficient by calculating the proportion of pipeline topology matching degree gradients that meet a threshold, and then selects a corresponding number of agents for analysis to obtain the pipeline maintenance matching degree. This matching degree reflects the potential increase in heat load under the current maintenance conditions. Finally, it combines the average pipeline topology matching degree and the pipeline maintenance matching degree to calculate a comprehensive pipeline matching degree, which is used to correct the first heat station load, resulting in a more accurate heat station load.

[0018] Through the above technical solutions, this application not only considers the basic factors of the building itself and the meteorological environment, but also analyzes the impact of pipeline topology and maintenance on the load. Through machine learning models and multi-dimensional matching degree analysis, it effectively improves the accuracy of load prediction for heating stations, provides a basis for the selection and efficient operation of integrated heating stations, and helps to reduce energy waste and ensure heating quality. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the integrated heat station load forecasting method based on machine learning provided in an embodiment of this application.

[0021] Figure 2 This is a schematic diagram of the structure of the integrated heating station load prediction system based on machine learning provided in the embodiments of this application.

[0022] The components represented by each number in the attached diagram are explained below:

[0023] Information acquisition module 11, model training module 12, data processing module 13, and load correction module 14. Detailed Implementation

[0024] This application provides an integrated heat station load forecasting method and system based on machine learning, which addresses the technical problem that existing heat station load forecasting methods are unable to fully consider the complex factors affecting heat load, resulting in inaccurate forecasting results.

[0025] Example 1, as Figure 1 As shown in the embodiments of this application, an integrated heat station load forecasting method based on machine learning is provided, including:

[0026] S10: Obtain the building characteristics and pipeline topology information of the target building to be assembled with the integrated heating station, and collect the historical meteorological data sequence of the environment in which the target building is located;

[0027] In this embodiment, firstly, the characteristics of the building and the pipes within the building used to connect to the heating station for heat transfer and return water are collected, such as the characteristics of the building's insulation, the area of ​​insulation, the building area, etc., and the pipe topology including pipe diameter, etc. Simultaneously, the temperature of the building environment over several past winters is collected to obtain historical meteorological data sequences.

[0028] Specifically, step S10 in the method includes:

[0029] Obtain the building features and pipe topology information of the target building to be equipped with the integrated heating station. The building features include building insulation parameters, and the pipe topology information includes the main pipe dimensions.

[0030] The average temperature of the environment where the target building is located over the past multiple heating cycles is collected to obtain a historical meteorological data sequence.

[0031] In this embodiment, firstly, the building characteristics of the target building to be equipped with the integrated heating station are obtained. These building characteristics include the building's insulation parameters, such as the thermal conductivity of the insulation material of the building's exterior walls, the thickness of the roof insulation layer, and the heat transfer coefficient of the exterior windows. Pipeline topology information records the heating pipe network structure inside the target building used to connect the integrated heating station, including the diameter, material, and length of the main pipes, the layout and number of branch pipes, and the connection method of each pipe node.

[0032] Secondly, for the environment of the target building, meteorological data was collected for multiple complete heating cycles in the past, such as the past 5 winters, specifically the daily average temperature. The daily average temperature was arranged in chronological order to construct a historical meteorological data sequence, which reflects the temperature change pattern of the environment of the target building during the heating season.

[0033] S20: Based on machine learning, according to the building characteristics and historical meteorological data sequence, perform load prediction for the first heating station, obtain the load of the first heating station, perform basic selection of integrated heating station, and obtain the first selection information;

[0034] In this embodiment of the application, the basic predicted heat load is first obtained based on machine learning to obtain the load of the first heat station, and then the integrated heat station is selected, that is, the model of the integrated heat station is selected to obtain the first selection information.

[0035] Specifically, based on machine learning, and according to the building characteristics and historical meteorological data sequences, the load of the first heating station is predicted to obtain the load of the first heating station, including:

[0036] Invoke a pre-configured, machine learning-based agent for predicting thermal loads.

[0037] Calculate the mean of the historical meteorological data series to obtain the historical average meteorological data;

[0038] The building features and historical average meteorological data are input into the heat load prediction agent, and the load of the first heat station is output.

[0039] The heat load prediction agent is built based on machine learning, using a set of sample building features and a set of sample historical average meteorological data as training inputs, and a set of sample heat station loads as supervised labels for training until test convergence. The set of sample building features, the set of sample historical average meteorological data, and the set of sample heat station loads are obtained based on the operation test data of integrated heat stations.

[0040] In this embodiment of the application, firstly, a heat load prediction agent pre-built and configured based on machine learning is invoked. This heat load prediction agent is trained with sample data. Its training input includes a set of sample building features and a set of sample historical average meteorological data. The supervision label is a set of sample heat station loads.

[0041] The sample data all came from the operational test data of actual integrated heating stations. Before making predictions, the collected historical meteorological data series were preprocessed, and their mean was calculated to obtain the historical average meteorological data, which was used as the representative value of meteorological factors.

[0042] Then, the building characteristics of the target building and the calculated historical average meteorological data are input into the heat load prediction agent. After internal neural network calculation and feature mapping, the agent outputs the load of the first heat station, which is a preliminary prediction result without considering pipeline topology and maintenance factors.

[0043] Furthermore, after obtaining the load of the first heating station, basic selection of integrated heating stations is performed to obtain initial selection information. Specifically, a preset heating station selection database is accessed, which stores information such as the rated load range, performance parameters, and applicable conditions of different models of integrated heating stations. The load of the first heating station is matched with the rated load range of each model of heating station in the database, and heating station models whose rated load can cover the load of the first heating station and whose other performance parameters initially meet the basic requirements of the target building are selected, forming the initial selection information. The initial selection information may include one or more candidate models.

[0044] For example, the following steps are taken to build and train a thermal load prediction agent based on machine learning:

[0045] First, data preparation involves collecting operational test data from integrated heating stations under different building types and meteorological conditions to construct a sample set. The sample building characteristic set includes multiple sets of insulation parameters for different buildings; the sample historical average meteorological data set represents the average temperature value of the environment where the sample building is located during the corresponding heating cycle; and the sample heating station load set consists of the measured load data of the integrated heating station in the sample building during actual operation.

[0046] Secondly, in terms of model construction, a long short-term memory network was selected as the basic model architecture. The number of nodes in the input layer of the model corresponds to the sum of the dimension of the building features and the number of historical average meteorological data. There are two hidden layers, each containing 64 neurons, and the output layer is one neuron, corresponding to the predicted heat station load value.

[0047] Next, for model training, the preprocessed set of sample building features and the sample historical average meteorological dataset were used as input, and the sample heat station load set was used as supervision labels, all input into the constructed LSTM model. The training was set to 100 epochs with a epoch size of 32, using the Adam optimizer with an initial learning rate of 0.001 and mean squared error as the loss function. During training, validation was performed every 10 epochs. The loss value on the validation set was used to determine if the model was overfitting. If the loss value on the validation set did not decrease for 5 consecutive epochs, training was stopped prematurely.

[0048] Finally, model evaluation and optimization are performed. After training, the model is evaluated using an independent test set, and the root mean square error (RMSE) and mean absolute percentage error (MAPE) between the predicted and measured loads are calculated. If RMSE and MAPE do not reach the preset thresholds, such as RMSE greater than 5% or MAPE greater than 3%, the number of hidden layer neurons, the learning rate, or the optimizer is changed, such as using the RMSprop optimizer. The model is then retrained until its evaluation metrics on the test set meet the preset requirements, thus completing the construction and training of the thermal load prediction agent.

[0049] Further, the basic selection of the integrated heating station is carried out to obtain the first selection information, including:

[0050] The system calls the heating station selection database, which stores the mapping relationship between heating station load ranges and heating station selection information.

[0051] The load of the first heating station is input into the heating station selection database. Based on the heating station load range it falls into, the corresponding heating station selection information is obtained as the first selection information.

[0052] In this embodiment, a pre-set heating station selection database is first invoked. This database pre-establishes a mapping relationship between heating station load ranges and heating station selection information. Each load range corresponds to one or more specific heating station models and their core parameters, such as rated heat output power, maximum flow rate, and inlet / outlet water temperature design values.

[0053] During the basic selection process, the load value of the first heat station obtained from the heat load prediction agent is input into the database. The database will automatically determine the preset load range into which the load of the first heat station falls, and then retrieve the heat station selection information corresponding to the load range according to the preset mapping relationship, as the first selection information for preliminary screening.

[0054] For example, if the load range in the heating station selection database is pre-defined as [50-100kW), [100-200kW), [200-300kW), etc., when the load of the first heating station is 150kW, it is determined that it falls into the [100-200kW) range. Then, the corresponding heating station model, such as "RHS-150" or "HST-180", is retrieved from the database, and the selection information such as the rated heat load, working pressure range, and external dimensions of the model are obtained simultaneously and integrated to form the first selection information.

[0055] S30: Obtain the baseline pipeline topology information group of the integrated heating station, perform topology matching degree analysis with the pipeline topology information, and obtain the pipeline topology matching degree gradient;

[0056] In this embodiment, pipeline topology information of similar models of heating stations with comparable operating loads and predicted loads over a past period is obtained as a baseline pipeline topology information group. Then, the similarity between the current pipeline topology information and the baseline pipeline topology information group is calculated to obtain the pipeline topology matching degree gradient, which includes multiple pipeline topology matching degrees.

[0057] Since integrated heating stations are prefabricated, there may be mismatches in the pipes within the building. Therefore, it is necessary to analyze the matching degree of pipe topology and pipe maintenance.

[0058] Specifically, step S30 in the method includes:

[0059] The integrated heating station with the first selection information is obtained from the historical operating data of the station, and filtered to obtain the reference pipeline topology information group and the reference pipeline stability coefficient sequence.

[0060] Calculate the similarity between the pipeline topology information and multiple reference pipeline topology information within the reference pipeline topology information group to obtain a pipeline topology similarity sequence;

[0061] The pipeline topology matching gradient is calculated based on the baseline pipeline stability coefficient sequence and the pipeline topology similarity sequence.

[0062] In this embodiment, firstly, for each candidate heating station model included in the first selection information, the operating data of that model of integrated heating station in multiple past practical application scenarios is retrieved from the historical operation database, and the operating data is filtered. For example, the filtering conditions include: the deviation between the operating load and the predicted load of the heating station is within a preset range of ±5%, the heating effect is stable, such as the user complaint rate is less than 0.5%, and no major failures have occurred in the pipeline system, etc.

[0063] Secondly, after screening, a set of reference pipeline topology information matching each candidate model is obtained, along with a sequence of reference pipeline stability coefficients within the corresponding operating cycle. The reference pipeline stability coefficient reflects the stability of the pipeline system during long-term operation and can be calculated comprehensively from indicators such as pipeline pressure fluctuation amplitude, flow stability, and heat loss rate. Its value ranges from 0 to 1, with values ​​closer to 1 indicating a more stable pipeline system.

[0064] Next, the similarity between the target building's pipe topology information and each baseline pipe topology information within the baseline pipe topology information group is calculated. Specifically, the pipe topology information is decomposed into multiple key feature dimensions, such as the main pipe diameter, the number of branch pipes, the thermal conductivity of the pipe material, the total pipe length, and the node connection method. For each feature dimension, normalization is applied to convert it into a value between 0 and 1. Then, the cosine similarity algorithm is used to calculate the similarity between the target pipe topology and each baseline pipe topology in the multi-dimensional feature space. That is, by converting each feature dimension into vector form, the similarity is calculated to obtain the pipe topology similarity sequence.

[0065] For example, the similarity between pipeline topology information and reference pipeline topology information is calculated. It is assumed that the main pipeline diameter of the target pipeline topology is DN150, the number of branch pipelines is 8, the thermal conductivity of the pipeline material is 0.025W / (m·K), the total pipeline length is 200 meters, and the node connection method is flange connection; the main pipeline diameter of a reference pipeline topology is DN150, the number of branch pipelines is 6, the thermal conductivity of the pipeline material is 0.024W / (m·K), the total pipeline length is 180 meters, and the node connection method is flange connection.

[0066] First, normalize each feature dimension: the main pipe diameter DN150 is normalized to 0.5 within the preset diameter range [DN100, DN200]; the number of branch pipes (8) is normalized to (8-4) / (12-4)=0.5 within the preset number range [4, 12], and the number of branch pipes (6) is normalized to (6-4) / 8=0.25; the thermal conductivity (0.025 W / (m·K)) is within the preset range [0.02, 0.02]. 03] is normalized to (0.025-0.02) / 0.01=0.5, 0.024 is normalized to 0.4; the total length of the pipeline is 200 meters, which is normalized to (200-100) / 200=0.5 within the preset range [100,300], and 180 meters is normalized to (180-100) / 200=0.4; if the node connection method is flange connection recorded as 1 and threaded connection recorded as 0, then both are 1.

[0067] Then, feature vectors are constructed, with the target pipeline topology vector being [0.5,0.5,0.5,0.5,1] and the baseline pipeline topology vector being [0.5,0.25,0.4,0.4,1].

[0068] Furthermore, the cosine similarity is used to calculate the similarity between the two. The cosine similarity is calculated as follows: Cosine similarity = (0.5×0.5+0.5×0.25+0.5×0.4+0.5×0.4+1×1) / [sqrt(0.5²+0.5²+0.5²+0.5²+1²)×sqrt(0.5²+0.25²+0.4²+0.4²+1²)] = 1.775 / [sqrt(2)×sqrt(1.6325)] ≈ 0.982, which gives the similarity value of the baseline pipeline topology. Here, sqrt is the arithmetic square root function. The above calculation is performed on all baseline pipeline topology information within the baseline pipeline topology information group to obtain the pipeline topology similarity sequence.

[0069] Finally, the pipeline topology matching gradient is calculated based on the baseline pipeline stability coefficient sequence and the pipeline topology similarity sequence. Specifically, for each element in the sequence, the pipeline topology similarity is multiplied by the corresponding baseline pipeline stability coefficient to obtain the matching degree of a single baseline pipeline topology. Then, all the single matching degrees are arranged in ascending or descending order to form the pipeline topology matching gradient.

[0070] Specifically, the integrated heating station that acquires the first selection information obtains historical operating data over a period of time, and filters it to obtain a reference pipeline topology information group and a reference pipeline stability coefficient sequence, including:

[0071] The integrated heating station that acquires the first selection information operates within a historical time period, and obtains a set of historical pipeline topology information, a set of historical predicted heating station loads, and a set of sample actual heating station loads after operation.

[0072] Calculate the similarity between the historical predicted heat station load set and the sample actual heat station load set to obtain the historical load consistency coefficient set;

[0073] Within the set of historical load consistency coefficients, the top K historical load consistency coefficients with the largest values ​​are selected as the baseline pipeline stability coefficient sequence, and the corresponding historical pipeline topology information is removed to obtain the baseline pipeline topology information group.

[0074] In this embodiment, firstly, from the historical operating data of the integrated heating station corresponding to the first selection information, a set of historical pipeline topology information, a set of historical predicted heating station loads, and a set of sample actual heating station loads recorded after operation are extracted. The set of historical pipeline topology information includes pipeline topology feature data under different historical application scenarios; the set of historical predicted heating station loads refers to the predicted heating station load values ​​obtained through the prediction model at the time of the corresponding scenario; and the set of sample actual heating station loads is the actual load data generated by the heating station during actual operation under the aforementioned scenarios.

[0075] Secondly, the similarity between the historical predicted load set and the sample actual load set is calculated to obtain the historical load consistency coefficient set. Specifically, for each set of historical predicted loads and sample actual loads, the formula "1 - |mean absolute percentage error|" is used to convert the accuracy of the load prediction into a coefficient characterizing the degree of consistency between the two. For example, if a set of historical predicted loads is 100kW, the sample actual load is 105kW, and the mean absolute percentage error is 4.76%, then the historical load consistency coefficient can be expressed as 1 - 0.0476 = 0.9524. The closer this coefficient is to 1, the higher the consistency between the historical predicted load and the actual load, meaning that the matching effect between the pipeline topology and the heat station load in this historical scenario may be better.

[0076] Subsequently, from the obtained set of historical load consistency coefficients, the top K historical load consistency coefficients with the largest values ​​are selected, where K is a preset positive integer, for example, K=20, which can be adjusted according to the actual data volume and model requirements. The selected coefficients are used as the baseline pipeline stability coefficient sequence. At the same time, based on the correspondence between the selected historical load consistency coefficients and the original data, the corresponding historical pipeline topology information is extracted, which together constitute the baseline pipeline topology information group.

[0077] Through the above screening process, it is ensured that the pipeline topology information in the baseline pipeline topology information group comes from scenarios where the historical predicted load and actual load are highly consistent and the operating effect is relatively stable.

[0078] S40: Obtain the pipeline maintenance information of the target building, perform maintenance matching degree analysis based on the pipeline topology matching degree gradient, obtain the pipeline maintenance matching degree, calculate the pipeline matching degree in combination with the pipeline topology matching degree gradient, and correct the load of the first heating station to obtain the heating station load.

[0079] In this embodiment, pipeline maintenance information includes data such as pipeline installation time, historical maintenance records, periodic inspection reports, and daily maintenance frequency. Based on the pipeline topology matching gradient, when performing maintenance matching degree analysis, the pipeline maintenance information is first quantified into multiple maintenance feature indicators. Then, for each baseline pipeline topology in the baseline pipeline topology information group, the aforementioned maintenance feature indicators within the corresponding historical operating cycle are extracted to form a baseline maintenance feature vector. The cosine similarity between the target building's pipeline maintenance feature vector and each baseline maintenance feature vector is calculated to obtain a maintenance similarity sequence.

[0080] Furthermore, the maintenance similarity sequence is multiplied by the corresponding individual matching degree in the pipeline topology matching degree gradient to obtain the pipeline maintenance matching degree that comprehensively considers the topology and maintenance status.

[0081] Subsequently, the pipe maintenance matching degrees corresponding to all baseline pipe topologies are weighted and averaged to obtain the final pipe matching degree. The load of the first heating station is then corrected based on the pipe matching degree to obtain the corrected heating station load.

[0082] This includes obtaining the pipeline maintenance information of the target building, performing maintenance matching degree analysis based on the pipeline matching gradient, and obtaining the pipeline maintenance matching degree, including:

[0083] Obtain the pipeline maintenance information of the target building, wherein the pipeline maintenance information includes total operating time and maintenance frequency;

[0084] An integrated maintenance matching analysis agent group is obtained, wherein the maintenance matching analysis agent group includes K maintenance matching analysis agents, each maintenance matching analysis agent is constructed based on machine learning, using a sample selection information set and a sample pipeline maintenance information set as training inputs, and using a sample pipeline maintenance matching degree set as supervision labels for training, and the training data of the K maintenance matching analysis agents are not completely the same;

[0085] Calculate the proportion of pipe topology matching degree within the pipe topology matching degree gradient that is greater than or equal to the preset pipe topology matching degree threshold, and obtain the calling coefficient;

[0086] Based on the calling coefficient and K, the number of calls is calculated. Within the maintenance matching analysis agent group, the maintenance matching analysis agent with the number of calls is called. The first selection information is combined with the pipeline maintenance information as input, and the average of the output results is calculated to obtain the pipeline maintenance matching degree. The pipeline maintenance matching degree includes the magnitude of the increase in heat load of the integrated heating station under the pipeline maintenance information.

[0087] In this embodiment of the application, firstly, pipeline maintenance information is extracted from the facility maintenance records of the target building, including the total operating time of the pipeline since its installation and the maintenance frequency within the operating cycle, such as quantitative indicators like monthly maintenance frequency, quarterly maintenance frequency, or annual maintenance frequency.

[0088] Secondly, an integrated swarm of maintenance matching analysis agents is constructed, consisting of K independently trained maintenance matching analysis agents. Each maintenance matching analysis agent is built based on a deep learning model, such as a multilayer perceptron or a long short-term memory network.

[0089] Specifically, the training process is as follows: The set of sample selection information of different integrated heating stations in history, including heating station model, rated parameters and corresponding sample pipeline maintenance information, including total operating time, maintenance frequency, etc., are used as input features, and the sample pipeline maintenance matching degree set is used as supervision label.

[0090] The sample pipeline maintenance matching degree, determined through expert evaluation, characterizes the pipeline system's adaptability to the heating station load under specific maintenance information and the potential increase in load. For example, a sample pipeline with low maintenance frequency and long operating time corresponds to a 5% increase in heating load. The training data for each agent differs in the sample space, for example, through random sampling or dividing historical data into different time periods, to enhance the generalization ability and decision robustness of the agent swarm.

[0091] Next, calculate the call coefficient. First, preset the pipeline topology matching degree threshold, for example, 0.8. Iterate through all matching degree values ​​in the pipeline topology matching degree gradient, count the number of matching degrees greater than or equal to the threshold, and divide it by the total length of the matching degree gradient. The resulting ratio is the call coefficient. This call coefficient reflects the proportion of high-matching samples in the current pipeline topology and the baseline topology. The higher the coefficient, the richer the historical experience of high adaptability that can be referenced.

[0092] Then, based on the calling coefficient and the total number of agents K, the number of maintenance matching analysis agents to be called is calculated. For example, if the calling coefficient is 0.6 and K=5, then the number of calls = 0.6 × 5 = 3. Agents of this number are randomly selected from the maintenance matching analysis agent group or sorted by performance. The initial selection information and the pipeline maintenance information of the target building are integrated into an input vector, which is then input into each selected maintenance matching analysis agent. Each agent outputs a predicted pipeline maintenance matching degree, representing the extent to which the load of the first heating station may need to increase under the current maintenance information to ensure stable operation of the heating station, such as 3%, 4.5%, or 5%. The arithmetic mean of the output results yields the final pipeline maintenance matching degree.

[0093] Furthermore, by combining the pipeline topology matching degree gradient, the pipeline matching degree is calculated, and the load of the first heating station is corrected to obtain the heating station load, including:

[0094] Calculate the mean value of the pipeline topology matching degree gradient to obtain the average pipeline topology matching degree;

[0095] The pipeline matching degree is calculated based on the average pipeline topology matching degree and the pipeline maintenance matching degree.

[0096] Based on the pipeline matching degree, the load correction coefficient is calculated, and the load of the first heating station is corrected to obtain the heating station load for reference display.

[0097] In this embodiment, firstly, the mean of the pipeline topology matching degree gradient is calculated, that is, the arithmetic mean of all individual matching degree values ​​in the gradient is taken to obtain the average pipeline topology matching degree. For example, if the pipeline topology matching degree gradient contains 20 matching degree values, and their sum is 18.5, then the average pipeline topology matching degree is 18.5 / 20 = 0.925. This mean reflects the overall average similarity level between the target pipeline topology and the baseline pipeline topology information group.

[0098] Secondly, the pipe matching degree is calculated by weighted summation based on the average pipe topology matching degree and the pipe maintenance matching degree. The weight of the average pipe topology matching degree is preset to α, and the weight of the pipe maintenance matching degree is preset to β, with α + β = 1. The weights α and β can be determined based on actual engineering experience or through cross-validation. For example, considering the basic influence of pipe topology on load, α = 0.7 and β = 0.3 can be set. If the average pipe topology matching degree is 0.925 and the pipe maintenance matching degree is 0.88, then the pipe matching degree = 0.925 × 0.7 + 0.88 × 0.3 = 0.9115.

[0099] Then, the load correction factor is calculated based on the pipe matching degree. The load correction factor is positively correlated with the pipe matching degree. For example, the formula "load correction factor = 1 - (1 - pipe matching degree) × γ" can be used, where γ is a correction factor that can be calibrated based on the relationship between pipe matching degree and load deviation in historical data, for example, γ = 0.3. If the pipe matching degree is 0.9115, then the load correction factor = 1 - (1 - 0.9115) × 0.3 = 0.97345.

[0100] Finally, the load of the first heating station is multiplied by a load correction factor to obtain the corrected load of the heating station. For example, if the load of the first heating station is 1200kW, then the corrected load of the heating station = 1200 × 0.97345 ≈ 1168.14kW. This corrected load of the heating station is displayed in numerical or graphical form on the system interface for reference, allowing heating station operation and management personnel to formulate scheduling strategies or monitor the load.

[0101] In summary, compared with existing technologies, this application has constructed a complete technical link from basic load forecasting to multi-dimensional dynamic correction by deeply integrating the selection information of integrated heating stations, pipeline topology characteristics and maintenance conditions, which effectively improves the accuracy and adaptability of heating station load forecasting.

[0102] In summary, the embodiments of this application have at least the following technical effects:

[0103] This application provides an integrated heat station load prediction method based on machine learning. First, it acquires the building characteristics and pipeline topology information of the target building, as well as historical meteorological data sequences of the environment in which the target building is located. Second, it uses a pre-built and configured heat load prediction agent to output the first heat station load. This agent is trained on a large amount of sample data to ensure the model's predictive ability. Then, based on the first heat station load, it calls a heat station selection database to perform basic selection, obtaining first selection information and initially determining the range of heat station models. Further, it determines the calling coefficient by calculating the proportion of pipeline topology matching degree gradients that meet a threshold, and then selects a corresponding number of agents for analysis to obtain the pipeline maintenance matching degree. This matching degree reflects the potential increase in heat load under the current maintenance conditions. Finally, it combines the average pipeline topology matching degree and the pipeline maintenance matching degree to calculate a comprehensive pipeline matching degree, which is used to correct the first heat station load, resulting in a more accurate heat station load.

[0104] Through the above technical solutions, this application not only considers the basic factors of the building itself and the meteorological environment, but also analyzes the impact of pipeline topology and maintenance on the load. Through machine learning models and multi-dimensional matching degree analysis, it effectively improves the accuracy of load prediction for heating stations, provides a basis for the selection and efficient operation of integrated heating stations, and helps to reduce energy waste and ensure heating quality.

[0105] Example 2, as Figure 2 As shown, based on the same inventive concept as the machine learning-based integrated heating station load forecasting method provided in Embodiment 1, this application also provides a machine learning-based integrated heating station load forecasting system, including:

[0106] The information acquisition module 11 is used to acquire the building characteristics and pipeline topology information of the target building to be assembled with the integrated heating station, as well as to acquire the historical meteorological data sequence of the environment in which the target building is located;

[0107] Model training module 12 is used to predict the load of the first heating station based on machine learning, according to the building characteristics and historical meteorological data sequence, to obtain the load of the first heating station, to perform integrated heating station basic selection, and to obtain the first selection information.

[0108] Data processing module 13 is used to acquire the reference pipeline topology information group of the integrated heating station, perform topology matching degree analysis with the pipeline topology information, and obtain the pipeline topology matching degree gradient.

[0109] The load correction module 14 is used to obtain the pipeline maintenance information of the target building, perform maintenance matching degree analysis based on the pipeline topology matching degree gradient, obtain the pipeline maintenance matching degree, calculate the pipeline matching degree in combination with the pipeline topology matching degree gradient, and correct the load of the first heating station to obtain the heating station load.

[0110] In one embodiment, the information acquisition module 11 is specifically used for:

[0111] Obtain the building features and pipe topology information of the target building to be equipped with the integrated heating station. The building features include building insulation parameters, and the pipe topology information includes the main pipe dimensions.

[0112] The average temperature of the environment where the target building is located over the past multiple heating cycles is collected to obtain a historical meteorological data sequence.

[0113] Further, in one embodiment of the application, based on machine learning, and according to the building characteristics and historical meteorological data sequences, the load of the first heating station is predicted to obtain the load of the first heating station, including:

[0114] Invoke a pre-configured, machine learning-based agent for predicting thermal loads.

[0115] Calculate the mean of the historical meteorological data series to obtain the historical average meteorological data;

[0116] The building features and historical average meteorological data are input into the heat load prediction agent, and the load of the first heat station is output.

[0117] Furthermore, in one embodiment of the application, the heat load prediction agent is constructed based on machine learning, using a sample building feature set and a sample historical average meteorological data set as training inputs, and a sample heat station load set as a supervised label for training until test convergence. The sample building feature set, the sample historical average meteorological data set, and the sample heat station load set are obtained based on the operational test data collection of integrated heat stations.

[0118] Further, the basic selection of the integrated heating station is carried out to obtain the first selection information, including:

[0119] The system calls the heating station selection database, which stores the mapping relationship between heating station load ranges and heating station selection information.

[0120] The load of the first heating station is input into the heating station selection database. Based on the heating station load range it falls into, the corresponding heating station selection information is obtained as the first selection information.

[0121] In one embodiment, the data processing module 13 is specifically used for:

[0122] The integrated heating station with the first selection information is obtained from the historical operating data of the station, and filtered to obtain the reference pipeline topology information group and the reference pipeline stability coefficient sequence.

[0123] Calculate the similarity between the pipeline topology information and multiple reference pipeline topology information within the reference pipeline topology information group to obtain a pipeline topology similarity sequence;

[0124] The pipeline topology matching gradient is calculated based on the baseline pipeline stability coefficient sequence and the pipeline topology similarity sequence.

[0125] Furthermore, in one embodiment, the integrated heating station acquiring the first selection information obtains operational data over a historical period, and filters it to obtain a reference pipeline topology information group and a reference pipeline stability coefficient sequence, including:

[0126] The integrated heating station that acquires the first selection information operates within a historical time period, and obtains a set of historical pipeline topology information, a set of historical predicted heating station loads, and a set of sample actual heating station loads after operation.

[0127] Calculate the similarity between the historical predicted heat station load set and the sample actual heat station load set to obtain the historical load consistency coefficient set;

[0128] Within the set of historical load consistency coefficients, the top K historical load consistency coefficients with the largest values ​​are selected as the baseline pipeline stability coefficient sequence, and the corresponding historical pipeline topology information is removed to obtain the baseline pipeline topology information group.

[0129] Further, in one embodiment, the pipe maintenance information of the target building is obtained, and based on the pipe matching gradient, a maintenance matching degree analysis is performed to obtain the pipe maintenance matching degree, including:

[0130] Obtain the pipeline maintenance information of the target building, wherein the pipeline maintenance information includes total operating time and maintenance frequency;

[0131] An integrated maintenance matching analysis agent group is obtained, wherein the maintenance matching analysis agent group includes K maintenance matching analysis agents, each maintenance matching analysis agent is constructed based on machine learning, using a sample selection information set and a sample pipeline maintenance information set as training inputs, and using a sample pipeline maintenance matching degree set as supervision labels for training, and the training data of the K maintenance matching analysis agents are not completely the same;

[0132] Calculate the proportion of pipe topology matching degree within the pipe topology matching degree gradient that is greater than or equal to the preset pipe topology matching degree threshold, and obtain the calling coefficient;

[0133] Based on the calling coefficient and K, the number of calls is calculated. Within the maintenance matching analysis agent group, the maintenance matching analysis agent with the number of calls is called. The first selection information is combined with the pipeline maintenance information as input, and the average of the output results is calculated to obtain the pipeline maintenance matching degree. The pipeline maintenance matching degree includes the magnitude of the increase in heat load of the integrated heating station under the pipeline maintenance information.

[0134] Furthermore, by combining the pipeline topology matching degree gradient, the pipeline matching degree is calculated, and the load of the first heating station is corrected to obtain the heating station load, including:

[0135] Calculate the mean value of the pipeline topology matching degree gradient to obtain the average pipeline topology matching degree;

[0136] The pipeline matching degree is calculated based on the average pipeline topology matching degree and the pipeline maintenance matching degree.

[0137] Based on the pipeline matching degree, the load correction coefficient is calculated, and the load of the first heating station is corrected to obtain the heating station load for reference display.

Claims

1. An integrated heating station load forecasting method based on machine learning, characterized in that, The method includes: Acquire the building characteristics and pipeline topology information of the target building to be assembled with the integrated heating station, and collect the historical meteorological data sequence of the environment in which the target building is located; Based on machine learning, the load of the first heating station is predicted according to the building characteristics and historical meteorological data sequence, the load of the first heating station is obtained, and the basic selection of the integrated heating station is carried out to obtain the first selection information. Obtain the baseline pipeline topology information group of the integrated heating station, perform topology matching degree analysis with the pipeline topology information, and obtain the pipeline topology matching degree gradient; Obtain the pipeline maintenance information of the target building, perform maintenance matching degree analysis based on the pipeline topology matching degree gradient to obtain the pipeline maintenance matching degree, calculate the pipeline matching degree in combination with the pipeline topology matching degree gradient, and correct the load of the first heating station to obtain the heating station load.

2. The integrated heating station load forecasting method based on machine learning according to claim 1, characterized in that, Acquire the building characteristics and pipeline topology information of the target building to be equipped with the integrated heating station, and collect historical meteorological data sequences of the environment in which the target building is located, including: Obtain the building features and pipe topology information of the target building to be equipped with the integrated heating station. The building features include building insulation parameters, and the pipe topology information includes the main pipe dimensions. The average temperature of the environment where the target building is located over the past multiple heating cycles is collected to obtain a historical meteorological data sequence.

3. The integrated heating station load forecasting method based on machine learning according to claim 1, characterized in that, Based on machine learning, and according to the building characteristics and historical meteorological data sequences, the load of the first heating station is predicted to obtain the load of the first heating station, including: Invoke a pre-configured, machine learning-based agent for predicting thermal loads. Calculate the mean of the historical meteorological data series to obtain the historical average meteorological data; The building features and historical average meteorological data are input into the heat load prediction agent, and the load of the first heat station is output.

4. The integrated heating station load forecasting method based on machine learning according to claim 3, characterized in that, The heat load prediction agent is built based on machine learning. It uses a set of sample building features and a set of sample historical average meteorological data as training inputs, and a set of sample heat station loads as supervision labels to train until test convergence. The set of sample building features, the set of sample historical average meteorological data, and the set of sample heat station loads are obtained based on the operation test data of integrated heat stations.

5. The integrated heating station load forecasting method based on machine learning according to claim 1, characterized in that, Conduct basic selection for integrated heating stations to obtain initial selection information, including: The system calls the heating station selection database, which stores the mapping relationship between heating station load ranges and heating station selection information. The load of the first heating station is input into the heating station selection database. Based on the heating station load range it falls into, the corresponding heating station selection information is obtained as the first selection information.

6. The integrated heating station load forecasting method based on machine learning according to claim 1, characterized in that, Obtain the baseline pipeline topology information group of the integrated heating station, perform topology matching degree analysis with the pipeline topology information, and obtain the pipeline topology matching degree gradient, including: The integrated heating station with the first selection information is obtained from the historical operating data of the station, and filtered to obtain the reference pipeline topology information group and the reference pipeline stability coefficient sequence. Calculate the similarity between the pipeline topology information and multiple reference pipeline topology information within the reference pipeline topology information group to obtain a pipeline topology similarity sequence; The pipeline topology matching gradient is calculated based on the baseline pipeline stability coefficient sequence and the pipeline topology similarity sequence.

7. The integrated heating station load forecasting method based on machine learning according to claim 6, characterized in that, The integrated heating station that provides the first selection information has historical operating data, which is then filtered to obtain a baseline pipeline topology information group and a baseline pipeline stability coefficient sequence, including: The integrated heating station that acquires the first selection information operates within a historical time period, and obtains a set of historical pipeline topology information, a set of historical predicted heating station loads, and a set of sample actual heating station loads after operation. Calculate the similarity between the historical predicted heat station load set and the sample actual heat station load set to obtain the historical load consistency coefficient set; Within the set of historical load consistency coefficients, the top K historical load consistency coefficients with the largest values ​​are selected as the baseline pipeline stability coefficient sequence, and the corresponding historical pipeline topology information is removed to obtain the baseline pipeline topology information group.

8. The integrated heating station load forecasting method based on machine learning according to claim 1, characterized in that, Obtain the pipeline maintenance information of the target building, perform maintenance matching degree analysis based on the pipeline matching gradient, and obtain the pipeline maintenance matching degree, including: Obtain the pipeline maintenance information of the target building, wherein the pipeline maintenance information includes total operating time and maintenance frequency; An integrated maintenance matching analysis agent group is obtained, wherein the maintenance matching analysis agent group includes K maintenance matching analysis agents, each maintenance matching analysis agent is constructed based on machine learning, using a sample selection information set and a sample pipeline maintenance information set as training inputs, and using a sample pipeline maintenance matching degree set as supervision labels for training, and the training data of the K maintenance matching analysis agents are not completely the same; Calculate the proportion of pipe topology matching degree within the pipe topology matching degree gradient that is greater than or equal to the preset pipe topology matching degree threshold, and obtain the calling coefficient; Based on the calling coefficient and K, the number of calls is calculated. Within the maintenance matching analysis agent group, the maintenance matching analysis agent with the number of calls is called. The first selection information is combined with the pipeline maintenance information as input, and the average of the output results is calculated to obtain the pipeline maintenance matching degree. The pipeline maintenance matching degree includes the magnitude of the increase in heat load of the integrated heating station under the pipeline maintenance information.

9. The integrated heating station load forecasting method based on machine learning according to claim 1, characterized in that, Based on the pipeline topology matching gradient, the pipeline matching degree is calculated, and the load of the first heating station is corrected to obtain the heating station load, including: Calculate the mean value of the pipeline topology matching degree gradient to obtain the average pipeline topology matching degree; The pipeline matching degree is calculated based on the average pipeline topology matching degree and the pipeline maintenance matching degree. Based on the pipeline matching degree, a load correction coefficient is calculated to correct the load of the first heating station, and the heating station load is obtained for reference display.

10. An integrated heating station load forecasting system based on machine learning, characterized in that, The method for performing the integrated heat station load forecasting method based on machine learning as described in any one of claims 1-9 includes: The information acquisition module is used to acquire the building characteristics and pipeline topology information of the target building to be assembled with the integrated heating station, as well as to collect the historical meteorological data sequence of the environment in which the target building is located; The model training module is used to predict the load of the first heating station based on machine learning, according to the building characteristics and historical meteorological data sequence, to obtain the load of the first heating station, to perform integrated heating station basic selection, and to obtain the first selection information. The data processing module is used to acquire the baseline pipeline topology information group of the integrated heating station, perform topology matching degree analysis with the pipeline topology information, and obtain the pipeline topology matching degree gradient. The load correction module is used to acquire the pipeline maintenance information of the target building, perform maintenance matching degree analysis based on the pipeline topology matching degree gradient, obtain the pipeline maintenance matching degree, calculate the pipeline matching degree in combination with the pipeline topology matching degree gradient, and correct the load of the first heating station to obtain the heating station load.

Images