Heat supply network management metering and charging system based on Internet and data processing

Abstract

The invention relates to the technical field of heat supply network management, discloses a heat supply network management metering and charging system based on the Internet and data processing, and aims to overcome the defects of a traditional heat supply network management system in the aspects of data acquisition, energy consumption prediction, charging processing, anomaly detection and the like. The system comprises a data acquisition unit for collecting heat supply network operation data in real time; the energy consumption prediction unit is used for predicting future user energy consumption by using a time sequence data set training model; the charging unit is used for generating a hot fee bill in combination with the prediction result and the user type; the anomaly detection unit identifies data anomaly through a model and ensures stable operation of the heat supply network; and the system optimization unit is used for adjusting the operation parameters according to the abnormal detection result and optimizing the heat supply network efficiency. The advanced data processing technology and the internet communication means are adopted, efficient, accurate and intelligent heat supply network management is achieved, the accuracy of energy consumption prediction is improved, the fairness of charging processing is enhanced, and safe and stable operation of the heat supply network is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of heat network management, and particularly to a heat network management metering and charging system based on the Internet and data processing. Background Art

[0002] With the continuous development and improvement of the urban central heating system, heat network management is facing increasingly complex challenges. Traditional heat network management metering and charging systems have become difficult to meet the needs of modern heat network management in some aspects and have many limitations.

[0003] In terms of the billing model, traditional billing methods are mostly based on fixed billing standards and historical heat consumption data of users. This billing model ignores the actual energy consumption situation of users and the differences in user types, resulting in a lack of fairness in the billing results. For users with low energy consumption, they may need to pay higher fees that do not match their actual energy consumption; while for users with high energy consumption, they may lack the motivation to save energy due to unreasonable billing standards. This billing method not only fails to stimulate users' awareness of energy conservation but also is not conducive to the sustainable development of the heat network and the efficient utilization of energy.

[0004] In terms of anomaly detection, traditional heat network management systems rely on manual inspections or threshold judgments to identify and handle abnormal situations. However, abnormal situations during the operation of the heat network are often complex and diverse. Manual inspections are difficult to achieve real-time monitoring and comprehensive coverage, and simple threshold judgments may cause false alarms or missed alarms due to improper settings. This anomaly detection method not only affects the safe and stable operation of the heat network but also may lead to energy waste, equipment damage, and economic losses. Summary of the Invention

[0005] The purpose of the present invention is to provide a heat network management metering and charging system based on the Internet and data processing to solve the problems raised in the above background art.

[0006] To achieve the above purpose, the present invention provides the following technical solution: A heat network management metering and charging system based on the Internet and data processing, the system includes:

[0007] A data collection unit for collecting temperature data, flow data, pressure data, and user heat consumption data in the heat network in real time;

[0008] An energy consumption prediction unit that constructs an energy consumption prediction model. The input data of the energy consumption prediction model is the data collected by the data collection unit, and the output data is the predicted energy consumption of each user in a future period of time; the model training uses a time series data set containing historical temperature, flow, pressure, and heat consumption data, and the corresponding label is the actual energy consumption.

[0009] Billing unit, constructing a billing model, where the input data of the billing model is the prediction result of the energy consumption prediction model and the user type, and the output data is the user's heating fee bill; the model is trained using a combined dataset containing historical energy consumption and user type, and the corresponding label is the actual heating fee bill amount;

[0010] Anomaly detection unit, constructing an anomaly detection model, where the input data of the anomaly detection model is the data collected by the data collection unit, and the output data is the anomaly detection result, including normal data and abnormal data; the model is trained using a dataset containing normal data and known abnormal data, and the corresponding label is the data status category;

[0011] System optimization unit, connected to the anomaly detection unit, adjusting the heat network operation parameters according to the anomaly detection result to optimize the heat network efficiency.

[0012] Preferably, the energy consumption prediction model is constructed using the long short-term memory network (LSTM) algorithm. The long-term dependencies in time series data are captured through LSTM units. Its network structure includes:

[0013] Input layer: Receiving preprocessed temperature, flow rate, pressure, and heat consumption data;

[0014] LSTM layer: Containing multiple LSTM units, each unit processes one time step in the sequence data and passes the cell state and hidden state to the next unit;

[0015] Fully connected layer: Mapping the hidden state output by the LSTM layer to the predicted energy consumption through the fully connected layer;

[0016] Output layer: Outputting the predicted energy consumption of each user for a future period of time.

[0017] Preferably, the steps for constructing the billing model include:

[0018] Step A: Using the energy consumption prediction result and user type as input features, and the actual heating fee bill amount as the target variable;

[0019] Step B: Selecting the radial basis function kernel as the kernel function;

[0020] Step C: Training the model using the support vector machine algorithm to find the optimal hyperplane to distinguish the heating fee bill amounts of different user types;

[0021] Step D: For new input data, using the trained model for prediction and outputting the user's heating fee bill.

[0022] Preferably, the optimization steps of the billing model include:

[0023] Step 1: Adjusting the parameters of the support vector machine, including the penalty parameter C and the kernel function parameters;

[0024] Step 2: Use grid search combined with cross-validation to evaluate the model performance and select the optimal parameter combination;

[0025] Step 3: Use the validation set to validate the optimized model and evaluate the accuracy and mean squared error (MSE) of the model;

[0026] Step 4: When the model performance reaches the preset standard, save the model parameters to obtain the trained billing model.

[0027] Preferably, the anomaly detection model is constructed using the isolation forest algorithm. The specific method is as follows:

[0028] Randomly select eigenvalue and split value to construct a binary tree, so that the abnormal data is isolated to the shallower layer of the tree;

[0029] Form an isolation forest by constructing multiple isolation trees, and each tree independently performs anomaly detection on the data;

[0030] For the new input data, calculate its path length in each tree. The shorter the path, the more abnormal the data;

[0031] Integrate the path length results of multiple trees to determine whether the data is abnormal.

[0032] Preferably, the training steps of the anomaly detection model include:

[0033] Step I: Randomly select subset features and subset data to construct an isolation tree;

[0034] Step II: Repeat Step I to construct multiple isolation trees to form an isolation forest;

[0035] Step III: Set the anomaly score threshold, calculate the anomaly score according to the path length of the data in the isolation forest, and if it exceeds the threshold, it is determined as abnormal.

[0036] Preferably, the implementation method of the system optimization module includes:

[0037] Construct a parameter adjustment strategy library: Pre-define multiple heat network operation parameter adjustment strategies. Each strategy corresponds to different anomaly types and degrees of anomaly, and the strategy content covers the direction and amplitude of parameter adjustment;

[0038] Construct a decision engine: Receive the detection results of the anomaly detection unit, and match the corresponding parameter adjustment strategy in the parameter adjustment strategy library according to the anomaly type and degree of anomaly;

[0039] Construct an execution unit: Adjust the heat network operation parameters according to the parameter adjustment strategy matched by the decision engine to optimize the heat network efficiency.

[0040] Preferably, the specific method for constructing the parameter adjustment strategy library is as follows:

[0041] Set a rule table as the data structure for storing parameter adjustment strategies. Each rule contains a condition set and an action set. The condition set is used to record the exception types and degrees required to trigger the rule, and the action set is used to define the specific strategies for parameter adjustment.

[0042] Encode the exception types and degrees. Use a predefined enumeration type to represent the exception types and continuous numerical values to represent the exception degrees.

[0043] The rule table adopts a hash table structure to efficiently store and retrieve rule information.

[0044] Preferably, the system optimization module further includes a rule matcher, which has a condition parser built in. Using the condition parser, the rule matcher quickly parses specific conditions in the condition set, and then finds the rules in the rule table that match the encoded exception types and degrees, and executes the corresponding parameter adjustment strategies.

[0045] Preferably, the data acquisition unit uses the MQTT protocol to achieve real-time transmission of data to the cloud server for storage and processing.

[0046] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0047] Through the energy consumption prediction unit, this system collects and analyzes the temperature, flow, pressure and user heat consumption data in the heat network in real time, and constructs an accurate energy consumption prediction model. This model fully considers the actual energy consumption situation of users and the differences in user types, making the billing result more in line with the actual energy consumption level of users. The billing unit bills based on the prediction results of the energy consumption prediction model and user types, avoiding the problem of unfair billing caused by fixed billing standards and historical heat consumption data in traditional billing methods. For users with low energy consumption, they can pay fees that match their actual energy consumption; for users with high energy consumption, they can also be incentivized to save energy through reasonable billing standards.

[0048] This system is equipped with an anomaly detection unit, which constructs an anomaly detection model to monitor the abnormal conditions in the operation of the heat network in real time. This model can accurately identify complex and diverse abnormal conditions, avoiding false alarms or missed alarms caused by traditional manual inspections or simple threshold judgments. The timely response of the anomaly detection model can ensure the safe and stable operation of the heat network, effectively preventing energy waste, equipment damage and economic losses caused by abnormal conditions.

[0049] The system optimization unit is closely connected to the anomaly detection unit, and adjusts the operation parameters of the heat network in a timely manner according to the anomaly detection results to optimize the heat network efficiency. This mechanism enables the heat network to operate more efficiently, improve energy utilization efficiency, and reduce operating costs. Through system optimization, the heat network can better adapt to the changing needs of different users, enhancing service quality and user satisfaction. Brief Description of the Drawings

[0050] Figure 1 It is the working principle diagram of the heat network management metering and charging system described in the present invention;

[0051] Figure 2 It is the flow chart of the construction and optimization steps of the billing model and the construction principle of the anomaly detection model;

[0052] Figure 3 It is the working flow chart from data selection, isolation forest construction, anomaly detection to system optimization. Detailed Embodiments

[0053] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0054] Please refer to Figures 1-3 , the present invention provides a technical solution: a heat network management metering and charging system based on the Internet and data processing, and the system includes:

[0055] Data acquisition unit: responsible for collecting various key data in the heat network in real time, including but not limited to temperature data, flow data, pressure data, and user heat consumption data. To achieve this goal, the data acquisition unit is equipped with high-precision sensors and automation devices, which are deployed at key nodes of the heat network, such as heat sources, heat exchange stations, pipelines, and user ports. The sensors collect data at a preset frequency (such as every minute or every second) and transmit the data to the central server of the data acquisition unit through wired or wireless communication methods. The central server performs preliminary processing on the data, such as denoising, calibration, and formatting, to ensure the accuracy and consistency of the data, providing a reliable basis for subsequent data analysis and model training.

[0056] Energy consumption prediction unit: Responsible for constructing an energy consumption prediction model. The input data of this model comes from the real-time data collected by the data acquisition unit, including the time series of historical temperature, flow rate, pressure, and heat consumption data. The output of the model is the predicted energy consumption of each user in a future period (such as hours, days, or weeks). During the model construction process, it is necessary to preprocess the historical data, including steps such as data cleaning, missing value filling, and normalization. Machine learning algorithms are used to train the preprocessed data. During the training process, a time series data set containing historical energy consumption as labels is used to optimize the model's parameters to improve the prediction accuracy. After training is completed, the model can predict the future energy consumption based on real-time input data, providing decision support for the operation scheduling and energy management of the heat network.

[0057] Billing unit: Responsible for constructing a billing model to generate heat fee bills based on the actual energy consumption of users and user types. The input data of the billing model includes the prediction results of the energy consumption prediction model and user types (such as residential users, commercial users, or industrial users, etc.). The output of the model is the amount of the heat fee bill for the user. During the billing model construction process, it is necessary to combine historical energy consumption and user types to form a training data set. Machine learning techniques are used to fit the training data set to establish the mapping relationship between energy consumption and the amount of the heat fee bill. During the training process, the actual heat fee bill amount is used as a label to optimize the model's parameters. After training is completed, the model can generate accurate heat fee bills based on the energy consumption prediction results and user types of users, improving the fairness and rationality of billing.

[0058] Abnormal detection unit: Responsible for constructing an abnormal detection model to monitor abnormal situations during the operation of the heat network in real time. The input data of this model comes from the real-time data collected by the data acquisition unit, including temperature, flow rate, pressure, and heat consumption data, etc. The output of the model is the abnormal detection result, including normal data and abnormal data. During the abnormal detection model construction process, it is necessary to collect a data set containing normal data and known abnormal data, and label the data status category (normal or abnormal) for each data sample. The training data set is used for training to learn the feature distributions of normal data and abnormal data. After training is completed, the model can monitor the operation status of the heat network based on real-time input data, and promptly detect and handle abnormal situations to ensure the safe and stable operation of the heat network.

[0059] System Optimization Unit: Connected to the anomaly detection unit, it is responsible for adjusting the operation parameters of the heat network according to the anomaly detection results to optimize the heat network efficiency. When the anomaly detection model detects an anomaly, the system optimization unit immediately starts the optimization process. It conducts a detailed analysis of the anomaly data to determine the cause and location of the anomaly. According to the type and severity of the anomaly, corresponding adjustment strategies are formulated, such as adjusting the heat source output power, changing the pipeline flow rate or pressure, etc. The adjustment strategy is sent to the control system of the heat network, and the control system executes the adjustment operation to restore the normal operation state of the heat network and improve the efficiency.

[0060] The present invention will be further described below in conjunction with Embodiments 1 to 4:

[0061] Embodiment 1:

[0062] The energy consumption prediction model is constructed using the Long Short-Term Memory (LSTM) algorithm to capture the long-term dependencies in time series data, thereby accurately predicting the energy consumption of each user in the next period of time. The network structure of the energy consumption prediction model mainly includes an input layer, an LSTM layer, a fully connected layer, and an output layer.

[0063] The input layer is responsible for receiving the preprocessed temperature, flow rate, pressure, and heat consumption data. These data are sourced from the data acquisition unit and, after preprocessing steps such as cleaning, missing value filling, and normalization, form a time series data set. The data at each time step includes temperature values, flow rate values, pressure values, and heat consumption values, which serve as the input features of the LSTM model. For example, assuming the data acquisition unit collects data once per minute, the data for one day constitutes a time series, which contains 1440 time steps (24 hours * 60 minutes). The data at each time step is organized into a feature vector, such as [temperature, flow rate, pressure, heat consumption], and used as the input to the input layer.

[0064] The LSTM layer contains multiple LSTM cells, and each cell processes one time step in the sequence data. The LSTM cells transmit information through the cell state and the hidden state. The cell state is used to preserve long-term memory, and the hidden state is used to represent the output of the current time step. In specific implementation, the number of LSTM cells in the LSTM layer can be set as needed to capture different features in the data. For example, 128 LSTM cells can be set. Each cell processes the feature vector passed from the input layer and updates its cell state and hidden state. These states are passed to the next LSTM cell to form the long-term dependencies of the time series.

[0065] The fully connected layer maps the hidden state output by the LSTM layer to the predicted energy consumption. The hidden state output by the last unit of the LSTM layer contains information about the entire time series. The fully connected layer converts this hidden state into a predicted value through a linear transformation or a non-linear activation function (such as ReLU, Sigmoid, etc.). For example, the fully connected layer can be a linear layer with a single neuron, and its weights and biases are learned through the training process. The output of this neuron is the predicted energy consumption, representing the predicted energy consumption value of a certain user in the future for a period of time.

[0066] The output layer is responsible for outputting the predicted energy consumption of each user in the future for a period of time. According to specific application requirements, the output layer can output prediction results at different time granularities, such as hourly, daily, or weekly energy consumption predictions. During the training process, a time series dataset containing historical energy consumption as labels is used to optimize the parameters of the LSTM model. Through optimization methods such as backpropagation algorithm and gradient descent, the weights of the LSTM layer, the weights and biases of the fully connected layer are continuously adjusted to minimize the prediction error.

[0067] After training is completed, the energy consumption prediction model can predict the energy consumption of each user in the future for a period of time based on real-time input temperature, flow rate, pressure, and heat consumption data. This prediction ability is of great significance for the management and scheduling of the heat network, which can help operators plan energy distribution in advance, optimize the operation efficiency of the heat network, and reduce energy waste.

[0068] Embodiment 2:

[0069] The billing model is constructed based on the support vector machine (SVM) algorithm, aiming to accurately predict the amount of the heat fee bill of users according to the energy consumption prediction results and user types. The construction steps of the billing model include:

[0070] Step A: Use the energy consumption prediction results (such as the predicted energy consumption of users in the future for a period of time) and user types (such as residential users, commercial users, industrial users, etc.) as input features. Use the actual heat fee bill amount as the target variable, that is, the output that the model needs to predict. For example, for each user, there is a corresponding energy consumption prediction result and user type, and there is also an actual heat fee bill amount. These data will be used to train the billing model.

[0071] Step B: Select the radial basis function (RBF) kernel as the kernel function. The RBF kernel is a commonly used kernel function that can handle non-linear data and has good generalization ability.

[0072] Step C: Use the support vector machine algorithm to train the model. The SVM algorithm finds the optimal hyperplane to distinguish data points of different categories, that is, to distinguish the heat fee bill amounts of different user types here.

[0073] During the training process, the algorithm attempts to find a hyperplane that maximally separates the heat consumption bill amounts of different user types in the feature space.

[0074] Step D: For the new input data (i.e., the new energy consumption prediction results and user types), use the trained model for prediction and output the heat consumption bill amount of the user.

[0075] The optimization steps of the billing model include:

[0076] Step 1: Adjust the parameters of the support vector machine, including the penalty parameter C and the kernel function parameter (such as the γ parameter of the RBF kernel). The penalty parameter C controls the degree of penalty for misclassification. The larger the C value, the heavier the penalty for misclassification; the γ parameter controls the width of the RBF kernel, affecting the complexity and generalization ability of the model.

[0077] Step 2: Adopt the grid search method to search for the optimal parameter combination within the parameter space. Combine cross-validation to evaluate the model performance, and use k-fold cross-validation (such as k = 5) to reduce the risk of overfitting. In each cross-validation, divide the dataset into k subsets, take turns training the model with k - 1 subsets, and use the remaining 1 subset to verify the model performance.

[0078] Step 3: Use the validation set to verify the optimized model and evaluate the accuracy and mean squared error (MSE) of the model. The accuracy reflects the proportion of correct predictions of the model, and the MSE reflects the average squared error between the predicted values and the actual values of the model.

[0079] Step 4: When the model performance reaches the preset standard (such as the accuracy reaching a certain level and the MSE being small), save the model parameters to obtain the trained billing model. The trained model can be used for actual heat consumption bill prediction, providing an accurate billing basis for the heat network management system.

[0080] Example 3:

[0081] This example is used to describe the construction method of an anomaly detection model. The model is based on the isolation forest algorithm, aiming to construct a binary tree by randomly selecting feature values and splitting values, making it easier to isolate abnormal data to the shallower layers of the tree, thereby achieving effective detection of abnormal data. The algorithm principle of the anomaly detection model:

[0082] Isolated Tree Construction: Randomly select eigenvalue and splitting value to construct a binary tree. During the construction process, for each node, randomly select a feature and randomly select a splitting value within the range of the feature's values, and divide the data into left and right child nodes according to the splitting value. This process is carried out recursively until the stopping condition is met (such as reaching the maximum depth of the tree, the amount of data in the node is less than a certain threshold, etc.). Since abnormal data is usually sparse in the feature space, they are more likely to be isolated in the shallower layers of the tree.

[0083] Isolated Forest Formation: Form an isolated forest by constructing multiple isolated trees. Each tree independently performs anomaly detection on the data, improving the stability and accuracy of the model. Each tree in the isolated forest is independently constructed, and they are trained using different subsets of features and subsets of data, thus increasing the diversity of the model.

[0084] Anomaly Detection: For new input data, calculate its path length in each tree. The path length refers to the number of edges that the data passes through from the root node to the leaf node. The shorter the path, the more abnormal the data. Because abnormal data is more likely to be isolated in the shallower layers of the tree, their path lengths are usually shorter. Combine the path length results of multiple trees to determine whether the data is abnormal. It can be achieved by calculating the anomaly score of the data. The higher the anomaly score, the more abnormal the data.

[0085] The training steps of this model include:

[0086] Step I: Isolated Tree Construction: Randomly select subset features and subset data. Subset features are a part of features randomly selected from all features, and subset data is a part of data randomly sampled from the entire dataset. Use the selected subset features and subset data to construct an isolated tree. According to the construction principle of the isolated tree, recursively select eigenvalue and splitting value until the stopping condition is met.

[0087] Step II: Isolated Forest Construction: Repeat Step I to construct multiple isolated trees. Each tree is trained using different subsets of features and subsets of data to increase the diversity of the model. Combine the constructed multiple isolated trees into an isolated forest. Each tree in the isolated forest independently performs anomaly detection on the data.

[0088] Step III: Abnormal Score Threshold Setting and Abnormal Judgment: Set the abnormal score threshold. The abnormal score is calculated based on the path length of the data in the isolation forest, and the threshold can be set by statistically analyzing the distribution of the path length or based on experience. For new input data, calculate its abnormal score in the isolation forest. The abnormal score can be defined as the average or weighted average of the path lengths of the data in all trees. Compare the abnormal score with the threshold, and if it exceeds the threshold, it is determined to be abnormal. If the abnormal score of the data is higher than the threshold, the data is considered abnormal; otherwise, the data is considered normal.

[0089] Example 4:

[0090] The system optimization module aims to optimize the heat network efficiency by dynamically adjusting the heat network operation parameters. The system optimization module includes the construction of a parameter adjustment strategy library, a decision-making engine, and an execution unit.

[0091] The implementation method of the system optimization module includes:

[0092] Construct a parameter adjustment strategy library: The rule table is used as the data structure for storing parameter adjustment strategies. Each rule contains a set of conditions and a set of actions. The set of conditions is used to record the types and degrees of abnormalities required to trigger the rule and is the basis for rule matching. The set of actions is used to define the specific strategies for parameter adjustment, including the direction of adjustment (such as increase or decrease) and the amplitude (such as the specific value or percentage of adjustment).

[0093] The types of abnormalities are represented by a predefined enumeration type. For example, they can be defined as "too high temperature", "too low pressure", etc. The degree of abnormality is represented by a continuous numerical value, such as a range or a specific value, to reflect the severity of the abnormality.

[0094] The rule table uses a hash table structure to efficiently store and retrieve rule information. The hash table maps the set of conditions to a location in the table through a hash function, thereby achieving fast lookup. Each rule has a unique key in the hash table, which is composed of the encoding combination of the type of abnormality and the degree of abnormality.

[0095] Construct a decision-making engine: The decision-making engine receives the detection results of the abnormal detection unit, which include the type and degree of abnormality. Based on the type and degree of abnormality, the decision-making engine matches the corresponding parameter adjustment strategy in the parameter adjustment strategy library. The matching process is implemented by a rule matcher, which searches for the corresponding rule in the rule table according to the encoding of the type and degree of abnormality.

[0096] Construction Execution Unit: The execution unit adjusts the operating parameters of the heat network according to the parameter adjustment strategy matched by the decision-making engine. The adjustment process can be automatic or manually executed after being confirmed by the operator. The execution unit is connected to the heat network control system and can adjust the operating parameters of the heat network in real time to optimize the heat network efficiency.

[0097] The specific method for constructing the parameter adjustment strategy library is as follows:

[0098] According to the actual requirements and experience of heat network operation, define multiple parameter adjustment strategies. Each strategy corresponds to different abnormal types and degrees of abnormality. The strategy content covers the direction and amplitude of parameter adjustment. Encode the abnormal types and degrees of abnormality for storage and retrieval in the rule table. The abnormal types are represented by an enumeration type, and the degrees of abnormality are represented by continuous numerical values. Use a hash table structure to store the rule table. Each rule has a unique key, which is composed of the encoding combination of the abnormal type and the degree of abnormality. The hash table provides efficient storage and retrieval functions and can quickly find the rules that match the abnormal type and the degree of abnormality.

[0099] The rule matcher has a condition parser built-in for quickly parsing specific conditions in the condition set. The condition parser can recognize the encodings of the abnormal type and the degree of abnormality and match them with the keys in the rule table. The rule matcher uses the condition parser to find the rules in the rule table that match the encodings of the abnormal type and the degree of abnormality. The search process is fast and efficient because the hash table structure provides fast key-value search functions. Once a matching rule is found, the rule matcher will execute the corresponding action set of the rule, that is, the specific strategy for adjusting the operating parameters of the heat network. The adjustment strategy is implemented through the execution unit to optimize the heat network efficiency.

[0100] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variation thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.

[0101] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A heat network management metering and charging system based on the Internet and data processing, characterized in that: include: Data acquisition unit, used to collect temperature data, flow data, pressure data and user heat data in the heating network in real time; An energy consumption prediction unit is used to construct an energy consumption prediction model, wherein the input data of the energy consumption prediction model is the data collected by the data acquisition unit, and the output data is the predicted energy consumption of each user in a future period of time; The model training uses a time series dataset containing historical temperature, flow, pressure, and heat usage data, with the corresponding label being the actual energy consumption; The billing unit constructs a billing model, wherein the input data of the billing model is the prediction result of the energy consumption prediction model and the user type, and the output data is the user's heating bill; the model training uses a combined data set containing historical energy consumption and user type, and the corresponding label is the actual heating bill amount; An anomaly detection unit constructs an anomaly detection model, wherein the input data of the anomaly detection model is the data collected by the data acquisition unit, and the output data is the anomaly detection result, including normal data and abnormal data; the model training uses a data set containing normal data and known abnormal data, and the corresponding label is the data state category; The system optimization unit is connected to the anomaly detection unit and adjusts the operation parameters of the heating network according to the anomaly detection result to optimize the efficiency of the heating network.

2. The heat network management metering and charging system based on the Internet and data processing according to claim 1 is characterized in that: The energy consumption prediction model is constructed using the long short-term memory network LSTM algorithm, which captures the long-term dependencies in time series data through LSTM units. Its network structure includes: Input layer: receives pre-processed temperature, flow, pressure and heat consumption data; LSTM layer: contains multiple LSTM units, each of which processes a time step in the sequence data and passes the cell state and hidden state to the next unit; Fully connected layer: maps the hidden state output by the LSTM layer to the predicted energy consumption through the fully connected layer; Output layer: Output the predicted energy consumption of each user in the future.

3. The heat network management metering and charging system based on the Internet and data processing according to claim 1 is characterized in that: The steps of constructing the billing model include: Step A: The energy consumption prediction results and user type are used as input features, and the actual heating bill amount is used as the target variable; Step B: Select the radial basis function kernel as the kernel function; Step C: Use the support vector machine algorithm to train the model and find the optimal hyperplane to distinguish the heating bill amounts of different user types; Step D: Use the trained model to make predictions on the new input data and output the user’s heating bill.

4. The heat network management metering and charging system based on the Internet and data processing according to claim 3 is characterized in that: The optimization steps of the billing model include: Step 1: Adjust the parameters of the support vector machine, including the penalty parameter C and the kernel function parameters; Step 2: Use grid search combined with cross-validation to evaluate model performance and select the optimal parameter combination; Step 3: Use the validation set to validate the optimized model and evaluate the accuracy and mean square error (MSE) of the model; Step 4: When the model performance reaches the preset standard, save the model parameters and obtain the trained billing model.

5. The heat network management metering and charging system based on the Internet and data processing according to claim 1 is characterized in that: The anomaly detection model is constructed using the isolation forest algorithm, and the specific method is as follows: Randomly select eigenvalues ​​and split values ​​to construct a binary tree so that abnormal data is isolated to the shallower layers of the tree; An isolation forest is formed by constructing multiple isolation trees, and each tree independently detects anomalies in the data; For new input data, calculate the path length in each tree. The shorter the path, the more abnormal the data. The path length results of multiple trees are combined to determine whether the data is abnormal.

6. The heat network management metering and charging system based on the Internet and data processing according to claim 5 is characterized in that: The training steps of the anomaly detection model include: Step I: Randomly select subset features and subset data to build an isolation tree; Step II: Repeat step I to construct multiple isolated trees to form an isolation forest; Step III: Set the anomaly score threshold and calculate the anomaly score based on the path length of the data in the isolation forest. If it exceeds the threshold, it is considered an anomaly.

7. The heat network management metering and charging system based on the Internet and data processing according to claim 1 is characterized in that: The implementation method of the system optimization module includes: Build a parameter adjustment strategy library: pre-define a variety of heat network operation parameter adjustment strategies, each strategy corresponds to different abnormality types and abnormality degrees, and the strategy content covers the direction and amplitude of parameter adjustment; Build a decision engine: Receive the detection results of the anomaly detection unit, and match the corresponding parameter adjustment strategy in the parameter adjustment strategy library according to the anomaly type and degree; Build execution unit: adjust the heating network operating parameters according to the parameter adjustment strategy matched by the decision engine to optimize the heating network efficiency.

8. The heat network management metering and charging system based on the Internet and data processing according to claim 7 is characterized in that: The specific method of building a parameter adjustment strategy library is: A rule table is set as a data structure for storing parameter adjustment strategies, where each rule contains a condition set and an action set. The condition set is used to record the type and degree of anomalies required to trigger the rule, and the action set is used to define the specific strategy for parameter adjustment. Encode the abnormality type and degree, use a predefined enumeration type to represent the abnormality type, and use a continuous value to represent the degree of abnormality; The rule table uses a hash table structure to efficiently store and retrieve rule information.

9. A heat network management metering and charging system based on the Internet and data processing according to claim 8, characterized in that: The system optimization module also includes a rule matcher, which has a built-in condition parser. Using the condition parser, the rule matcher quickly parses specific conditions in the condition set, and then finds the rules that match the abnormal type and abnormal degree coding in the rule table, and executes the corresponding parameter adjustment strategy.

10. The heat network management metering and charging system based on the Internet and data processing according to claim 1 is characterized in that: The data acquisition unit uses the MQTT protocol to achieve real-time transmission of data to the cloud server for storage and processing.

Images