Dynamic game-based cogeneration system optimization scheduling method and system

Abstract

The invention discloses a cogeneration system optimization scheduling method and system based on a dynamic game, and relates to the technical field of cogeneration scheduling, and the method comprises the steps: obtaining the information of a cogeneration system, obtaining the historical data of the cogeneration system according to the information of the cogeneration system, obtaining the user information and the user basic data, and carrying out the optimization scheduling of the cogeneration system based on the historical data of the cogeneration system; and classifying the users according to the user basic data to obtain user classification information. According to the method, the electricity consumption characteristic days corresponding to each user are obtained through the user basic data set, the thermoelectric demand characteristics of the users are accurately analyzed through the electricity consumption characteristic days, data clustering is carried out on the users through the strategy characteristic vectors, the user characteristics and behavior modes are deeply analyzed, a data basis is provided for subsequent system scheduling, and the user experience is improved. Through the user classification information, the thermoelectric demand corresponding to each type of users is accurately evaluated, and the accuracy and timeliness of the scheduling of the combined heat and power generation system are ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of combined heat and power scheduling, in particular to a combined heat and power system optimization scheduling method and system based on dynamic game. BACKGROUND

[0002] As an important carrier for realizing energy efficient utilization, combined heat and power systems are facing the key demand for transformation from traditional centralized to digital, intelligent and interactive. Combined heat and power systems improve energy utilization rate by simultaneously producing electricity and heat, and are the core link of building comprehensive energy systems. However, with the large-scale access of distributed energy, energy storage devices and multi-user terminals, the complexity of the system increases exponentially, and the diversity of user behavior and the asymmetry of information interaction become the main bottleneck restricting system optimization. The rapid development of digital technology provides a new path for the upgrading of combined heat and power systems. By collecting and analyzing real-time power and heat data, the operation of combined heat and power systems is optimized by means of dynamic game theory, and the intelligent and efficient level of energy systems is improved.

[0003] Currently, the optimization scheduling of combined heat and power systems cannot accurately analyze user data, cannot accurately evaluate user heat and power demand according to user data, and cannot timely and accurately schedule the heat and power system, reducing the stability and reliability of the heat and power system. In the prior art, all user data is often directly analyzed to evaluate user demand, but in actual process, the heat and power demand of different users often differs, and the presence of only one abnormal user can greatly affect the data analysis of the entire group, making it impossible to accurately evaluate user data. If each user is analyzed separately, not only is the data processing amount large, but also the data processing efficiency is low, reducing the scheduling efficiency of the combined heat and power system and affecting the timeliness and accuracy of the combined heat and power system scheduling. SUMMARY

[0004] To solve the above technical problems, a combined heat and power system optimization scheduling method and system based on dynamic game are provided. The present technical solution solves the problem of being unable to accurately analyze user data, being unable to accurately evaluate user heat and power demand according to user data, being unable to timely and accurately schedule the heat and power system, reducing the stability and reliability of the heat and power system, and the problem of being unable to accurately evaluate user data in the prior art, in which all user data is often directly analyzed to evaluate user demand. In actual process, the heat and power demand of different users often differs, and the presence of only one abnormal user can greatly affect the data analysis of the entire group. If each user is analyzed separately, not only is the data processing amount large, but also the data processing efficiency is low, reducing the scheduling efficiency of the combined heat and power system and affecting the timeliness and accuracy of the combined heat and power system scheduling.

[0005] To achieve the above object, the technical scheme adopted by the present application is: A method for optimizing scheduling of a combined heat and power system based on dynamic game, comprising: Obtaining combined heat and power system information, including combined heat and power system function information and combined heat and power system performance parameter information; According to the combined heat and power system information, obtaining combined heat and power historical data, including historical power supply data and historical heat supply data; Obtaining user information and user basic data, including historical power consumption data, user historical heat consumption data and user demand response data corresponding to each user; Based on the combined heat and power historical data, classifying users according to the user basic data to obtain user classification information, including main group user information and island user information; Obtaining user real-time load data, including user real-time power consumption load data and user real-time heat supply load data; According to the user classification information, taking the sum of the user real-time power consumption load corresponding to the main group users as the reference power supply load, and taking the sum of the user real-time heat supply load corresponding to the main group users as the reference heat supply load; According to the user basic data corresponding to the main group users, training the existing model to obtain a group demand prediction model; Based on the reference power supply load and the reference heat supply load, and based on the group demand prediction model, obtaining reference power supply prediction load and reference heat supply prediction load; According to the user basic data corresponding to the island users, obtaining the power supply demand deviation coefficient and the heat supply demand deviation coefficient corresponding to each island user, the power supply demand deviation coefficient representing the ratio of the average daily power consumption of the island user to the maximum daily power consumption, and the heat supply demand deviation coefficient representing the ratio of the average daily heat supply of the island user to the maximum daily heat supply; Taking the sum of the user real-time power consumption load corresponding to the island users as the characteristic power supply load, and taking the sum of the user real-time heat supply load corresponding to the island users as the characteristic heat supply load; Taking the product of the characteristic power supply load and the power supply demand deviation coefficient as the second power supply prediction load, and taking the product of the characteristic heat supply load and the heat supply demand deviation coefficient as the second heat supply prediction load; Taking the reference power supply prediction load and the second power supply prediction load as the total power supply demand load, and taking the reference heat supply prediction load and the second heat supply prediction load as the total heat supply demand load; According to the total power supply demand load and the total heat supply demand load, scheduling the combined heat and power system.

[0006] Preferably, the user is classified according to the user basic data based on the combined heat and power historical data, and user classification information is obtained, specifically comprising: According to the combined heat and power historical data and the user basic data, the user basic data corresponding to the same user is divided into the same data set, and the user basic data set is obtained; According to the user basic data set, the electricity characteristic day corresponding to each user is obtained; According to the user basic data set, the average daily electricity consumption, the peak-valley electricity consumption ratio and the electricity fluctuation coefficient in the electricity characteristic day of the user are obtained, the peak-valley electricity consumption ratio is the ratio of the total peak segment electricity consumption to the total valley segment electricity consumption, and the electricity fluctuation coefficient is the ratio of the difference between the maximum daily electricity consumption and the minimum daily electricity consumption to the average daily electricity consumption; According to the user basic data set and the electricity characteristic day, the average daily heat supply, the temperature adjustment frequency and the heat loss coefficient in the electricity characteristic day of the user are obtained, the temperature adjustment frequency is the number of daily temperature setting value changes of the user, and the heat loss coefficient is the ratio of the difference between the heat supply and the actual indoor heat absorption to the heat supply; According to the user demand response data and the combined heat and power historical data, the response participation rate, the response intensity and the response delay time are obtained based on the electricity characteristic day, the response participation rate is the ratio of the actual response times to the response invitation times, the response intensity is the ratio of the average load reduction amount of each response to the reference load, and the response delay time represents the average time length from the issuance of the response invitation to the start of the response; The user basic data is normalized by taking the average daily electricity consumption, the peak-valley electricity consumption ratio, the electricity fluctuation coefficient, the average daily heat supply, the temperature adjustment frequency, the heat loss coefficient, the response participation rate, the response intensity and the response delay time as characteristic dimensions, and the strategy characteristic vector corresponding to each user is obtained; The strategy characteristic vectors corresponding to the same user are divided into the same data set, and the strategy characteristic vector data set information is obtained; According to the strategy characteristic vector data set information, the data clustering matching algorithm is obtained based on data clustering characteristics analysis; Based on the data clustering matching algorithm, the data clustering of the strategy characteristic vector is performed, and the data clustering information is obtained; Based on the data clustering information, the user is classified, and the user classification information is obtained.

[0007] Preferably, the electricity characteristic day corresponding to each user is obtained according to the user basic data set, specifically comprising: According to the user basic data set, the historical electricity consumption data in each user basic data set is obtained, the historical electricity consumption data includes daily peak-valley period electricity consumption and daily total electricity consumption of the user; According to the daily peak and valley period electricity consumption of the user, the daily peak and valley period information of the user and the electricity consumption information corresponding to each peak and valley period are obtained; On the basis of the electricity consumption corresponding to any one user peak period in the daily peak and valley period of the user, the ratio of the electricity consumption in the valley period to the electricity consumption corresponding to the user peak period in the daily peak and valley period of the user is taken as the electricity consumption deviation coefficient of the user; The ratio of the minimum value of the electricity consumption deviation coefficient corresponding to the daily peak and valley period of the user to the maximum value of the electricity consumption deviation coefficient is taken as the electricity consumption stability difference coefficient corresponding to the user on the day; According to the electricity consumption stability difference coefficient and the user basic data set, the electricity consumption characteristic coefficient corresponding to each user is obtained; The electricity consumption characteristic coefficient is adjusted until the maximum value is reached, and the electricity consumption characteristic days corresponding to each user are obtained; The electricity consumption characteristic coefficient is adjusted until the maximum value is reached, and the electricity consumption characteristic days corresponding to each user are obtained; ; In the formula, is the electricity consumption characteristic coefficient, represents the maximum value of the electricity consumption stability difference coefficient in h days, represents the minimum value of the electricity consumption stability difference coefficient in h days, and h is the electricity consumption characteristic days.

[0008] Preferably, the data clustering matching algorithm is obtained based on data clustering feature analysis according to the strategy feature vector data set information, specifically including: According to the strategy feature vector data set information, whether each feature dimension in the strategy feature vector data set conforms to normal distribution is judged based on the KS test method, and the feature dimension conforming to the normal distribution is taken as the reference feature dimension; According to the reference feature dimension, the strategy feature vector corresponding to the reference feature dimension is taken as the reference feature vector, and the strategy feature vector which is not the reference feature vector is taken as the secondary feature vector; According to the reference feature vector, any two reference feature vectors are taken as a reference feature group; The sum of the Euclidean distances between the reference feature vector and all secondary feature vectors is taken as the feature divergence coefficient of the reference feature vector; The reference feature vector corresponding to the larger value of the feature divergence coefficient in the reference feature group is taken as the first feature vector, and the reference feature vector corresponding to the smaller value of the feature divergence coefficient is taken as the second feature vector; According to the reference feature group, the Pearson correlation coefficient between the two reference feature vectors in the reference feature group is obtained; According to the Pearson correlation coefficient, the reference feature vectors in the reference feature group are screened to obtain the calibration feature vector; If the absolute value of the Pearson correlation coefficient is less than 0.3, the benchmark feature vector in the benchmark feature group is a non-redundant feature, and if the absolute value of the Pearson correlation coefficient is not less than 0.3, the benchmark feature vector in the benchmark feature group is a redundant feature, and the second feature vector in the benchmark feature group is removed. According to the normal distribution coefficient, a data clustering matching algorithm is obtained. If the normal distribution coefficient is greater than 0.8, the K-means algorithm is used as the data clustering matching algorithm, and if the normal distribution coefficient is not greater than 0.8, the DBSCAN algorithm is used as the data clustering matching algorithm.

[0009] Preferably, the data clustering matching algorithm is used to perform data clustering on the strategy feature vector, specifically including: According to the data clustering matching algorithm, corresponding data clustering parameters are obtained. If the data clustering matching algorithm is the K-means algorithm, the elbow method is used to obtain the initial cluster number based on the strategy feature vector dataset information. Based on the K-means algorithm data clustering requirement, K-means iteration parameters are set, including the maximum iteration number, the cluster center convergence threshold, and the initial cluster center selection method. If the data clustering matching algorithm is the DBSCAN algorithm, the elbow method is used to obtain the field radius based on the strategy feature vector dataset information, and the minimum sample number is set. According to the data clustering parameters, the stability of the data clustering matching algorithm is verified, 70% of the strategy feature vector dataset is extracted based on stratified sampling as data samples, and the adjusted Rand index of each clustering result and the original result is calculated after repeating clustering for 5 times. According to the adjusted Rand index, it is determined whether the data clustering matching algorithm meets the data clustering requirement. If the adjusted Rand index is greater than 0.8, it is determined that the data clustering is stable, and if the adjusted Rand index is not greater than 0.8, it is determined that the data clustering is unstable, and the data clustering parameters are adjusted. If the data clustering matching algorithm is the K-means algorithm, the iteration number is increased or the initial center selection method is adjusted. If the data clustering matching algorithm is the K-means algorithm, the radius is reduced or the minimum sample number is reduced. According to the data clustering matching algorithm and the data clustering parameters, the strategy feature vector is data clustered to obtain data clustering information, including data cluster information, center strategy feature vector information corresponding to each data cluster, and cluster user number information.

[0010] Preferably, the data clustering information is used to classify users to obtain user classification information, specifically including: According to the data clustering information, center strategy feature vector information corresponding to each data cluster is obtained; The Euclidean distance of the center strategy feature vectors of any two data clusters is taken as the strategy difference degree of the two data clusters; According to the data clustering information and the strategy difference degree, a difference matrix D is constructed: ; In the formula, QUOTE is the strategy difference degree of the xth data cluster and the yth data cluster, is the strategy difference degree of the xth data cluster and the yth data cluster, According to the difference matrix, the mean and the standard deviation of all elements in the difference matrix are obtained; According to the mean and the standard deviation, a difference threshold is obtained; According to the user basic data set, data interaction frequencies corresponding to each data cluster are obtained, the data interaction frequency representing the number of times that the users in the cluster and the users in other clusters have consistent load adjustment directions in the same period; according to the data interaction frequencies corresponding to each data cluster, 1 / 3 of the average value of the data interaction frequencies corresponding to the data clusters is taken as a data interaction threshold; According to the difference threshold and the data interaction threshold, a strategy isolation index corresponding to each data cluster is obtained; Based on a preset strategy isolation index threshold and the strategy isolation index corresponding to each data cluster, it is judged whether the data cluster is an isolated data cluster, and user classification information is obtained; The difference threshold is specifically: ; In the formula, QUOTE is the difference threshold, is the mean of the difference matrix, is the standard deviation of the difference matrix; The strategy isolation index is specifically: ; In the formula, is the strategy isolation index of the xth data cluster, is the maximum value of the strategy difference degree of the xth data cluster and other data clusters, is the data interaction frequency of the xth data cluster, is the data interaction threshold, and is a weight coefficient, wherein .

[0011] Further, a combined heat and power system optimization scheduling system based on dynamic game is proposed, which is used to realize the optimization scheduling method as described above, and comprises: The main control module is configured to obtain corresponding data clustering parameters according to the data clustering matching algorithm, verify the stability of the data clustering matching algorithm according to the data clustering parameters, perform data clustering on the strategy feature vector according to the data clustering matching algorithm and the data clustering parameters, obtain data clustering information, classify users based on the data clustering information, obtain user classification information, train an existing model according to user basic data corresponding to the main group of users, obtain a group demand prediction model, obtain a benchmark power supply prediction load and a benchmark heating prediction load based on the group demand prediction model and based on the benchmark power supply load and the benchmark heating load, take the benchmark power supply prediction load and the second power supply prediction load as a total power supply demand load, take the benchmark heating prediction load and the second heating prediction load as a total heating demand load, and schedule the combined heat and power system according to the total power supply demand load and the total heating demand load. The information acquisition module is configured to obtain combined heat and power system information, including combined heat and power system function information and combined heat and power system performance parameter information, obtain combined heat and power historical data including historical power supply data and historical heating data according to the combined heat and power system information, obtain user information and user basic data including historical power consumption data, historical heating data and demand response data of each user corresponding to the user, and obtain real-time load data of the user including real-time power consumption load data and real-time heating load data of the user. The evaluation module is configured to divide user basic data corresponding to the same user into the same data set according to the combined heat and power historical data and the user basic data, obtain user basic data sets, obtain power consumption characteristic days corresponding to each user according to the user basic data sets, judge whether each feature dimension in the strategy feature vector data set conforms to a normal distribution based on the KS test method according to the strategy feature vector data set information, take the feature dimension conforming to the normal distribution as a benchmark feature dimension, and obtain the data clustering matching algorithm based on data clustering feature analysis based on the strategy feature vector data set information. The display module is configured to interact with the main control module and output and display user classification information, user real-time load data, benchmark power supply prediction load, benchmark heating prediction load, second power supply prediction load, second heating prediction load, total power supply demand load and total heating demand load.

[0012] Optionally, the main control module specifically comprises: The control unit is used for classifying users based on data clustering information, obtaining user classification information, training an existing model according to user basic data corresponding to a main group of users, obtaining a group demand prediction model, obtaining a benchmark power supply prediction load and a benchmark heating prediction load based on the benchmark power supply load and the benchmark heating load, obtaining a total power supply demand load by taking the benchmark power supply prediction load and a second power supply prediction load as the total power supply demand load and obtaining a total heating demand load by taking the benchmark heating prediction load and a second heating prediction load as the total heating demand load, and scheduling a combined heat and power system according to the total power supply demand load and the total heating demand load. The information receiving unit is used for receiving data and transmitting the data to the data processing unit. The data processing unit is used for obtaining corresponding data clustering parameters according to the data clustering matching algorithm, verifying the stability of the data clustering matching algorithm according to the data clustering parameters, performing data clustering on a strategy feature vector according to the data clustering matching algorithm and the data clustering parameters, and obtaining data clustering information.

[0013] Optionally, the information obtaining module specifically comprises: The first obtaining unit is used for obtaining combined heat and power system information, the combined heat and power system information including combined heat and power system function information and combined heat and power system performance parameter information, and obtaining combined heat and power historical data including historical power supply data and historical heating data according to the combined heat and power system information. The second obtaining unit is used for obtaining user information and user basic data, the user basic data including historical power consumption data, historical heating data and demand response data of each user, and obtaining real-time load data including real-time power consumption load data and real-time heating load data of the user.

[0014] Optionally, the evaluation module specifically comprises: The first evaluation unit is used for dividing user basic data corresponding to the same user into the same data set according to the combined heat and power historical data and the user basic data, obtaining user basic data sets, and obtaining power consumption feature days corresponding to each user according to the user basic data sets. The second evaluation unit is used for judging whether each feature dimension in the strategy feature vector data set conforms to a normal distribution based on the KS test method, taking the feature dimension conforming to the normal distribution as a benchmark feature dimension, and obtaining the data clustering matching algorithm based on the strategy feature vector data set information and data clustering feature analysis.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention proposes an optimized scheduling method and system for cogeneration systems based on dynamic game theory. By using a user-based dataset, the method obtains the number of days with electricity consumption characteristics for each user. This data allows for accurate analysis of the user's heat and power demand characteristics. Furthermore, by using strategy feature vectors to cluster user data, the method deeply analyzes user characteristics and behavioral patterns, providing a data foundation for subsequent system scheduling. Finally, by using user classification information, the method accurately assesses the heat and power demand corresponding to each user category, ensuring the accuracy and timeliness of cogeneration system scheduling. Attached Figure Description

[0016] Figure 1 Here is a flowchart of an optimized scheduling method for a combined heat and power system based on dynamic game theory proposed in this invention. Figure 2 This is a flowchart of the user classification information acquisition process in this invention; Figure 3 This is a flowchart of the process for obtaining the number of days of electricity consumption characteristics in this invention; Figure 4 This is a flowchart of the data clustering and matching algorithm in this invention; Figure 5 This is a block diagram of an optimized scheduling system for a combined heat and power system based on dynamic game theory, as proposed in this invention. Detailed Implementation

[0017] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0018] Reference Figure 1 - Figure 4 As shown in the figure, an optimal scheduling method for a combined heat and power system based on dynamic game theory in an embodiment of the present invention includes: Obtain information about the combined heat and power (CHP) system, including CHP system functional information and CHP system performance parameter information; Based on the information from the cogeneration system, historical cogeneration data is obtained, including historical power supply data and historical heating data. Acquire user information and basic user data, including historical electricity consumption data, historical heating data, and user demand response data for each user. Based on historical data of cogeneration, users are classified according to basic user data to obtain user classification information, which includes information on main group users and isolated user information. Specifically, based on the combined heat and power historical data, the user is classified according to the user basic data, and user classification information is obtained, specifically including: According to the combined heat and power historical data and the user basic data, the user basic data corresponding to the same user is divided into the same data set, and the user basic data set is obtained; According to the user basic data set, the electricity characteristic day corresponding to each user is obtained; According to the user basic data set, the average daily electricity consumption, the peak-valley electricity consumption ratio and the electricity fluctuation coefficient in the electricity characteristic day of the user are obtained, the peak-valley electricity consumption ratio is the ratio of the total peak segment electricity consumption and the total valley segment electricity consumption, and the electricity fluctuation coefficient is the ratio of the difference between the maximum daily electricity consumption and the minimum daily electricity consumption and the average daily electricity consumption; According to the user basic data set and the electricity characteristic day, the average daily heat supply, the temperature adjustment frequency and the heat loss coefficient in the electricity characteristic day of the user are obtained, the temperature adjustment frequency is the number of daily temperature setting value changes of the user, and the heat loss coefficient is the ratio of the difference between the heat supply and the actual indoor heat absorption and the heat supply; According to the user demand response data and the combined heat and power historical data, the response participation rate, the response intensity and the response delay time are obtained based on the electricity characteristic day, the response participation rate is the ratio of the actual response times and the response invitation times, the response intensity is the ratio of the average load reduction amount of each response and the reference load, and the response delay time represents the average time length from the response invitation to the start of response; The average daily electricity consumption, the peak-valley electricity consumption ratio, the electricity fluctuation coefficient, the average daily heat supply, the temperature adjustment frequency, the heat loss coefficient, the response participation rate, the response intensity and the response delay time are taken as characteristic dimensions to normalize the user basic data, and the strategy characteristic vector corresponding to each user is obtained; The strategy characteristic vector corresponding to the same user is divided into the same data set, and the strategy characteristic vector data set information is obtained; According to the strategy characteristic vector data set information, the data clustering matching algorithm is obtained based on data clustering characteristics analysis; Based on the data clustering matching algorithm, the data clustering of the strategy characteristic vector is performed, and the data clustering information is obtained; Based on the data clustering information, the user is classified, and the user classification information is obtained.

[0019] In the scheme, through multi-dimensional feature mining, accurate clustering and dynamic game adaptation, a user-side "accurate portrait-strategy coordination" foundation is established for the optimal dispatching of the combined heat and power system. Through multi-dimensional fusion, the energy characteristic differences of each user type are clearly distinguishable (e.g., a user cluster is "high electricity fluctuation + short response delay", which is suitable for undertaking emergency peak shaving tasks), solving the pain points of traditional classification "fuzzy behavior description". Through normalization processing, the dimension differences of feature quantities are eliminated (e.g., the units of electricity consumption and temperature regulation frequency are different), and then similar feature vectors are aggregated using clustering matching algorithms (e.g., K-means, DBSCAN), outputting user classification to provide differentiated targeting for dispatching strategies, avoiding resource waste caused by "one-size-fits-all" dispatching (e.g., excessive subsidies for high-response users and ineffective incentives for low-response users). Through multi-dimensional feature mining and data clustering, accurate quantification and classification of user-side energy behavior are realized, providing a clear mapping of "user type-strategy response" for the optimal dispatching of the combined heat and power system.

[0020] Specifically, according to the user basic data set, the electricity consumption characteristic days corresponding to each user are obtained, specifically including: According to the user basic data set, the historical electricity consumption data in the user basic data set is obtained, and the historical electricity consumption data includes daily peak-valley period electricity consumption and daily total electricity consumption of the user; According to the daily peak-valley period electricity consumption of the user, the daily peak-valley period information of the user and the electricity consumption information corresponding to each peak-valley period are obtained; Taking the electricity consumption corresponding to any user peak period in the daily peak-valley period of the user as the basis, the ratio of the valley period electricity consumption in the daily peak-valley period of the user to the electricity consumption corresponding to the user peak period is taken as the electricity consumption deviation coefficient of the user; The ratio of the minimum value of the electricity consumption deviation coefficient corresponding to the daily peak-valley period electricity consumption of the user to the maximum value of the electricity consumption deviation coefficient is taken as the electricity consumption stability difference coefficient corresponding to the user on that day; According to the electricity consumption stability difference coefficient and the user basic data set, the electricity consumption characteristic coefficient corresponding to each user is obtained; The electricity consumption characteristic coefficient is adjusted until the maximum value is reached, and the electricity consumption characteristic days corresponding to each user are obtained; Wherein, the electricity consumption characteristic coefficient is specifically: ; In the formula, is the electricity consumption characteristic coefficient, represents the maximum value of the electricity consumption stability difference coefficient within h days, represents the minimum value of the electricity consumption stability difference coefficient within h days, and h is the electricity consumption characteristic day.

[0021] In the scheme, the user electricity stability and characteristic days are accurately quantified to provide key user-side data support for cogeneration system optimization scheduling. Starting from the user electricity data in peak and valley periods, the electricity deviation coefficient and stable difference coefficient are calculated to construct the electricity characteristic coefficient formula. The user electricity behavior is analyzed in multiple dimensions to accurately identify the electricity stability / fluctuation characteristics (such as large peak-valley difference of factory users, and characteristic coefficient reflecting the fluctuation law), which makes up for the insufficient description of user electricity characteristics in traditional classification, adapts the scheduling strategy to the real electricity mode of users, and uses the electricity characteristic days as the key index of user characteristics to provide the core basis for subsequent clustering. Different characteristic days of users (such as long-term stable electricity of residents and short-term fluctuation of merchants), corresponding to different response costs and benefits in dynamic game scheduling, can optimize the strategy (such as designing flexible electricity price for users with short characteristic days and large fluctuations), improve the actual adaptability of game equilibrium solution, reduce supply-demand mismatch, and predict the load changes of different users in extreme weather and peak-valley periods based on the electricity law identified by characteristic days. When scheduling, the standby capacity is reserved in advance for high-fluctuation users reflected by characteristic days, and energy distribution is optimized for stable users to improve the system's ability to cope with uncertainty, enhance the flexibility and reliability of cogeneration systems, and help achieve dynamic balance between supply and demand.

[0022] Specifically, according to the strategy feature vector dataset information, based on data clustering feature analysis, the data clustering matching algorithm is obtained, specifically including: According to the strategy feature vector dataset information, based on KS test method, it is judged whether each feature dimension in the strategy feature vector dataset conforms to the normal distribution, and the feature dimension conforming to the normal distribution is taken as the reference feature dimension; According to the reference feature dimension, the strategy feature vector corresponding to the reference feature dimension is taken as the reference feature vector, and the strategy feature vector which is not the reference feature vector is taken as the secondary feature vector; According to the reference feature vector, any two reference feature vectors are taken as a reference feature group; The sum of the Euclidean distances between the reference feature vector and all secondary feature vectors is taken as the feature divergence coefficient of the reference feature vector; The reference feature vector corresponding to the larger value of the feature divergence coefficient in the reference feature group is taken as the first feature vector, and the reference feature vector corresponding to the smaller value of the feature divergence coefficient is taken as the second feature vector; According to the reference feature group, the Pearson correlation coefficient between the two reference feature vectors in the reference feature group is obtained; According to the Pearson correlation coefficient, the reference feature vectors in the reference feature group are screened to obtain the calibration feature vector; If the absolute value of the Pearson correlation coefficient is less than 0.3, the benchmark feature vector in the benchmark feature group is a non-redundant feature, and if the absolute value of the Pearson correlation coefficient is not less than 0.3, the benchmark feature vector in the benchmark feature group is a redundant feature, and the second feature vector in the benchmark feature group is removed. According to the normal distribution coefficient, a data clustering matching algorithm is obtained. If the normal distribution coefficient is greater than 0.8, the K-means algorithm is used as the data clustering matching algorithm, and if the normal distribution coefficient is not greater than 0.8, the DBSCAN algorithm is used as the data clustering matching algorithm.

[0023] In this scheme, a scientific clustering method is provided for user classification through feature screening and algorithm adaptation. The benchmark feature dimension based on KS test is identified, the redundant features (such as the strong correlation between “average daily electricity consumption” and “peak-valley electricity consumption ratio”) are removed based on the Pearson correlation coefficient (threshold 0.3), and the non-redundant benchmark feature vector is retained. This screening reduces the interference of noise features on clustering (such as removing the strong correlation features of heat loss coefficient and temperature regulation frequency), focuses the strategy feature vector on the core difference (such as demand response intensity and electricity fluctuation coefficient), improves the accuracy of subsequent clustering, and lays a high-quality data foundation for user classification. Based on the normal distribution coefficient (benchmark feature vector ratio), an adaptive algorithm is automatically selected: when the normal distribution feature ratio is more than 80%, K-means is used (suitable for spherical clusters and uniformly distributed data), otherwise DBSCAN is used (good at non-convex, non-uniform density clustering). For example, the electricity consumption characteristics of residential users are normally distributed (suitable for K-means), while the complex load characteristics of industrial parks are more suitable for DBSCAN. This “data-driven” algorithm selection avoids the classification bias caused by fixed algorithms, making the user clustering result more consistent with the actual energy consumption mode. The user classification output by the optimized clustering algorithm can more accurately reflect the demand response potential and energy consumption characteristics of different users.

[0024] Specifically, based on the data clustering matching algorithm, the strategy feature vector is data clustered, specifically including: According to the data clustering matching algorithm, the corresponding data clustering parameters are obtained; If the data clustering matching algorithm is the K-means algorithm, the initial cluster number is obtained based on the elbow method according to the strategy feature vector dataset information. Based on the data clustering requirement of the K-means algorithm, set the K-means iteration parameters, including the maximum number of iterations, the clustering center convergence threshold and the initial clustering center selection method; if the data clustering matching algorithm is the DBSCAN algorithm, then according to the strategy feature vector dataset information, based on the elbow method, obtain the domain radius, set the minimum sample number; according to the data clustering parameters, verify the stability of the data clustering matching algorithm, extract 70% from the strategy feature vector dataset based on stratified sampling as data samples, repeat clustering 5 times, and calculate the adjusted Rand index of each clustering result and the original result; According to the adjusted Rand index, determine whether the data clustering matching algorithm meets the data clustering requirement, if the adjusted Rand index is greater than 0.8, it is determined that the data clustering is stable, if the adjusted Rand index is not greater than 0.8, it is determined that the data clustering is unstable, and the data clustering parameters are adjusted; wherein, if the data clustering matching algorithm is the K-means algorithm, increase the number of iterations or adjust the initial center selection method; if the data clustering matching algorithm is the K-means algorithm, reduce the radius or reduce the minimum sample number; According to the data clustering matching algorithm and the data clustering parameters, the strategy feature vector is data clustered to obtain data clustering information, including data cluster information, center strategy feature vector information corresponding to each data cluster, and cluster user quantity information.

[0025] In this scheme, through accurate parameter configuration, stability verification and clustering optimization, reliable clustering results are provided for user grouping. According to the characteristic differences of K-means and DBSCAN algorithms, customized parameter configuration is made to adapt the distribution characteristics of strategy feature vectors (such as dense or sparse distribution of user electricity / heat features). For example, for residential user groups in the electricity feature set, the initial cluster number setting of K-means can accurately capture the "stable electricity" and "peak-valley fluctuation" subtypes; for industrial park users with scattered electricity features, the domain radius adjustment of DBSCAN can avoid misclassification of high / low load users into the same cluster. This parameter adaptation solves the clustering deviation caused by "one-size-fits-all" parameters, making the data clusters more consistent with the actual energy consumption patterns of users. Through stratified sampling, the adjusted Rand index (ARI) is calculated to verify the stability, and the parameters are dynamically adjusted to form a closed loop of "parameter setting → verification → optimization". This mechanism effectively avoids clustering distortion caused by accidental factors (such as sample fluctuations), ensuring the consistency of the clustering results. Stable clustering results are the basis for developing long-term scheduling strategies - avoiding repeated adjustments of strategies due to user grouping fluctuations, reducing scheduling costs, and through parameter adaptation, stability verification and result optimization, strategy feature vectors are converted into stable, accurate and interpretable user grouping information, providing a clear mapping of "user group characteristics → scheduling strategy" for dynamic game scheduling.

[0026] It should be noted that in this embodiment, the maximum number of iterations in the K-means iteration parameters is 100, the cluster center convergence threshold is 0.001 (i.e., the iteration stops when the distance between the centers of two iterations is less than 0.001), and the initial cluster center selection method uses the k-means++ algorithm to avoid random bias. Specifically, based on data clustering information, users are classified to obtain user classification information, which includes: Based on the data clustering information, obtain the central strategy feature vector information corresponding to each data cluster; The Euclidean distance between the central policy feature vectors of any two data clusters is taken as the policy difference between the two data clusters. Based on data clustering information and strategy dissimilarity, construct the dissimilarity matrix D: ; In the formula, QUOTE For the QUOTE The data cluster and the QUOTE The strategy variance of each data cluster; based on the variance matrix, obtain the mean and standard deviation of all elements in the variance matrix; Obtain the difference threshold based on the mean and standard deviation; Based on the user base dataset, the data interaction frequency corresponding to each data cluster is obtained. The data interaction frequency represents the number of times that the load adjustment direction of users in the same cluster is consistent with that of users in other clusters during the same period. Based on the data interaction frequency corresponding to each data cluster, 1 / 3 of the average data interaction frequency corresponding to the data cluster is used as the data interaction threshold. Based on the difference threshold and the data interaction threshold, obtain the strategy isolation index corresponding to each data cluster; Based on the preset policy isolation index threshold and the policy isolation index corresponding to each data cluster, determine whether the data cluster is an isolated data cluster and obtain user classification information; Specifically, the difference threshold is: ; In the formula, The difference threshold, The mean of the difference matrix, The standard deviation of the difference matrix; The isolation index of the strategy is specifically: ; In the formula, Let x be the policy isolation index for the x-th data cluster. a maximum strategy difference value of the xth data cluster and other data clusters, a data interaction frequency of the xth data cluster, a data interaction threshold value, a weight coefficient, wherein .

[0027] In the scheme, by constructing a quantitative model of strategy difference and interaction, the "island effect" of user groups is accurately identified, key support is provided for combined heat and power dispatching, the Euclidean distance of the strategy feature vector of the data cluster center is calculated, and a difference matrix is constructed. The difference threshold value is determined by the mean and standard deviation, the strategy difference of different user groups (such as the difference between "peak-valley power consumption type" and "stable power consumption type" groups) is quantified, the user group boundary is clearly divided, the relationship between different groups is more clear, and scheduling inefficiency caused by confusion is avoided. The data interaction frequency (intra-cluster and other cluster user load adjustment consistency) is introduced, and the strategy isolation index is calculated in combination with the difference threshold value. The "island data cluster" (such as a user group with large strategy difference and less interaction) is identified, which provides a basis for subsequent combined heat and power system dispatching and improves the efficiency and accuracy of dispatching.

[0028] It can be understood that in the combined heat and power system dispatching, the system is often dispatched according to the load demand of users. However, in the dispatching process, if the entire user group is directly taken as the basis for dispatching the system, the dispatching result obtained is often an equilibrium result. However, in the actual process, only a few abnormal users in the similar behavior user group will cause great changes in the dispatching result. If each user is analyzed separately, not only the data processing amount is large, but also it has a time delay, which cannot ensure the timeliness of the combined heat and power system.

[0029] It should be noted that in the embodiment, the strategy isolation index values of all clusters are sorted, the top 20% quantile value is taken as the strategy isolation index threshold value, the candidate island cluster with a strategy isolation index exceeding the threshold value is screened out, and the information interaction records of the users in the cluster (such as whether participating in the strategy discussion of the public information platform) are further checked. If more than 90% of the users have no cross-cluster information interaction, the potential island cluster is marked. The potential island cluster that simultaneously satisfies "there are at least three other clusters with a strategy difference exceeding the difference threshold value" and "the interaction frequency is lower than the data interaction threshold value" is determined as the island user group by comprehensively considering the difference and interaction between clusters.

[0030] Obtain real-time load data of users, wherein the real-time load data of users includes real-time power load data of users and real-time heating load data of users; According to the user classification information, the sum of the real-time power load of the users corresponding to the main group of users is taken as the reference power load, and the sum of the real-time heating load of the users corresponding to the main group of users is taken as the reference heating load; According to the user basic data corresponding to the main group user, the existing model is trained to obtain a group demand prediction model; Based on the benchmark power supply load and the benchmark heating load, the benchmark power supply prediction load and the benchmark heating prediction load are obtained based on the group demand prediction model; According to the user basic data corresponding to the island user, the power supply demand deviation coefficient and the heating demand deviation coefficient corresponding to each island user are obtained, the power supply demand deviation coefficient represents the ratio of the average daily power consumption of the island user to the maximum daily power consumption, and the heating demand deviation coefficient represents the ratio of the average daily heating amount of the island user to the maximum daily heating amount. The sum of the user real-time power consumption load corresponding to the island user is taken as the characteristic power supply load, and the sum of the user real-time heating load corresponding to the island user is taken as the characteristic heating load. The product of the characteristic power supply load and the power supply demand deviation coefficient is taken as the second power supply prediction load, and the product of the characteristic heating load and the heating demand deviation coefficient is taken as the second heating prediction load. The benchmark power supply prediction load and the second power supply prediction load are taken as the total power supply demand load, and the benchmark heating prediction load and the second heating prediction load are taken as the total heating demand load. According to the total power supply demand load and the total heating demand load, the combined heat and power system is dispatched.

[0031] It should be noted that based on the neural network model, the real-time demand of the user is predicted through the historical data, which is a common technical means known to those skilled in the art, so in this embodiment, it is not further expanded, and optionally, according to the user classification information, the behavior characteristic information corresponding to the main group user and the island user is obtained, and based on the existing dynamic game model, the combined heat and power system scheduling strategy is formulated.

[0032] Referring to Figure 5 Further, in combination with the above-mentioned one kind based on dynamic game's combined heat and power system optimization scheduling method, a kind of combined heat and power system optimization scheduling system based on dynamic game is proposed, including: The main control module is used for obtaining corresponding data clustering parameters according to a data clustering matching algorithm, verifying the stability of the data clustering matching algorithm according to the data clustering parameters, performing data clustering on a strategy feature vector according to the data clustering matching algorithm and the data clustering parameters, obtaining data clustering information, classifying users on the basis of the data clustering information, obtaining user classification information, training an existing model according to user basic data corresponding to a main group of users, obtaining a group demand prediction model, obtaining a benchmark power supply prediction load and a benchmark heating prediction load on the basis of the group demand prediction model, taking the benchmark power supply prediction load and a second power supply prediction load as a total power supply demand load, taking the benchmark heating prediction load and a second heating prediction load as a total heating demand load, and dispatching a combined heat and power system according to the total power supply demand load and the total heating demand load. The information acquisition module is used for acquiring combined heat and power system information, the combined heat and power system information including combined heat and power system function information and combined heat and power system performance parameter information, acquiring combined heat and power historical data according to the combined heat and power system information, the combined heat and power historical data including historical power supply data and historical heating data, acquiring user information and user basic data, the user basic data including historical power consumption data, historical heating data and demand response data of each user, and acquiring user real-time load data, the user real-time load data including user real-time power consumption load data and user real-time heating load data. The evaluation module is used for dividing user basic data of a same user into a same data set according to the combined heat and power historical data and the user basic data, acquiring user basic data sets, acquiring power consumption characteristic days of each user according to the user basic data sets, judging whether each characteristic dimension in a strategy feature vector data set conforms to a normal distribution based on a KS test method according to strategy feature vector data set information, taking a characteristic dimension conforming to the normal distribution as a benchmark characteristic dimension, and acquiring a data clustering matching algorithm based on data clustering feature analysis according to the strategy feature vector data set information. The display module interacts with the main control module and is used for outputting and displaying user classification information, user real-time load data, a benchmark power supply prediction load, a benchmark heating prediction load, a second power supply prediction load, a second heating prediction load, a total power supply demand load and a total heating demand load.

[0033] The main control module specifically includes: The control unit is used for classifying users based on data clustering information, obtaining user classification information, training an existing model according to user basic data corresponding to a main group of users, obtaining a group demand prediction model, obtaining a benchmark power supply prediction load and a benchmark heating prediction load based on the benchmark power supply load and the benchmark heating load, obtaining a total power supply demand load by taking the benchmark power supply prediction load and a second power supply prediction load as the total power supply demand load and obtaining a total heating demand load by taking the benchmark heating prediction load and a second heating prediction load as the total heating demand load, and scheduling a combined heat and power system according to the total power supply demand load and the total heating demand load. The information receiving unit is interactive with the information obtaining module and the evaluation module, and is used for receiving data and transmitting the data to the data processing unit. The data processing unit is used for obtaining corresponding data clustering parameters according to the data clustering matching algorithm, verifying the stability of the data clustering matching algorithm according to the data clustering parameters, and performing data clustering on the strategy feature vector according to the data clustering matching algorithm and the data clustering parameters to obtain data clustering information.

[0034] The information obtaining module specifically comprises: The first obtaining unit is used for obtaining combined heat and power system information, wherein the combined heat and power system information comprises combined heat and power system function information and combined heat and power system performance parameter information, and obtaining combined heat and power historical data according to the combined heat and power system information, wherein the combined heat and power historical data comprises historical power supply data and historical heating data. The second obtaining unit is used for obtaining user information and user basic data, wherein the user basic data comprises historical power consumption data, historical heating data and demand response data of each user corresponding to the user, and obtaining real-time load data of the user, wherein the real-time load data of the user comprises real-time power consumption load data and real-time heating load data of the user.

[0035] The evaluation module specifically comprises: The first evaluation unit is used for dividing user basic data corresponding to the same user into the same data set according to the combined heat and power historical data and the user basic data, obtaining the user basic data set, and obtaining power consumption feature days corresponding to each user according to the user basic data set. The second evaluation unit is used for judging whether each feature dimension in the strategy feature vector data set conforms to a normal distribution based on the KS test method, taking the feature dimension conforming to the normal distribution as a benchmark feature dimension, and obtaining the data clustering matching algorithm based on the strategy feature vector data set information and data clustering feature analysis.

[0036] In summary, the application has the advantages that: through the user base data set, the corresponding electricity characteristic days of each user are obtained, the thermal power demand characteristics of the user are accurately analyzed through the electricity characteristic days, the user is data clustered through the strategy characteristic vector, the user characteristics and behavior patterns are deeply analyzed, a data basis is provided for subsequent system scheduling, the thermal power demand corresponding to each type of user is accurately evaluated through the user classification information, and the accuracy and timeliness of the combined heat and power system scheduling are ensured.

[0037] The basic principles, main features and advantages of the application are shown and described above. Those skilled in the art should understand that the application is not limited by the above examples, and the above examples and descriptions in the specification are only the principles of the application. Without departing from the spirit and scope of the application, various changes and improvements can be made to the application, and these changes and improvements all fall within the scope of the claimed application. The scope of protection of the application is defined by the appended claims and their equivalents.

Claims

1. A method for optimal scheduling of a combined heat and power (CHP) system based on dynamic game theory, characterized in that, include: Obtain information about the combined heat and power (CHP) system, including CHP system functional information and CHP system performance parameter information; Based on the information from the cogeneration system, historical cogeneration data is obtained, including historical power supply data and historical heating data. Acquire user information and basic user data, including historical electricity consumption data, historical heating data, and user demand response data for each user. Based on historical data of cogeneration, users are classified according to basic user data to obtain user classification information, which includes information on main group users and isolated user information. Acquire real-time load data of users, including real-time electricity load data and real-time heating load data of users; Based on user classification information, the sum of real-time electricity loads of users corresponding to the main group of users is used as the benchmark power supply load, and the sum of real-time heating loads of users corresponding to the main group of users is used as the benchmark heating load. The existing model is trained based on the user base data corresponding to the main user group to obtain a group demand prediction model; Based on the benchmark power supply load and benchmark heating load, the benchmark power supply forecast load and benchmark heating forecast load are obtained based on the group demand forecast model. Based on the user base data corresponding to the isolated users, obtain the power supply demand deviation coefficient and heating demand deviation coefficient for each isolated user. The power supply demand deviation coefficient represents the ratio of the average daily electricity consumption of the isolated user to the maximum daily electricity consumption, and the heating demand deviation coefficient represents the ratio of the average daily heating supply of the isolated user to the maximum daily heating supply. The sum of the real-time electricity loads of the users corresponding to the isolated users is taken as the characteristic power supply load, and the sum of the real-time heating loads of the users corresponding to the isolated users is taken as the characteristic heating load. The product of the characteristic power supply load and the power supply demand deviation coefficient is used as the second power supply predicted load, and the product of the characteristic heating load and the heating demand deviation coefficient is used as the second heating predicted load. The baseline power supply forecast load and the second power supply forecast load are used as the total power supply demand load, and the baseline heating forecast load and the second heating forecast load are used as the total heating demand load. The combined heat and power (CHP) system is scheduled based on the total load demand for electricity and heat.

2. The method for optimal scheduling of a combined heat and power system based on dynamic game theory according to claim 1, characterized in that, The process of classifying users based on historical data from combined heat and power (CHP) and user base data to obtain user classification information specifically includes: Based on historical data of cogeneration and user basic data, the user basic data corresponding to the same user is divided into the same dataset to obtain the user basic dataset; Based on the user base dataset, obtain the number of days of electricity consumption characteristics for each user; Based on the user's basic dataset, the average daily electricity consumption, peak-valley electricity consumption ratio, and electricity fluctuation coefficient are obtained within the user's electricity consumption characteristic days. The peak-valley electricity consumption ratio is the ratio of the total electricity consumption during peak periods to the total electricity consumption during valley periods. The electricity fluctuation coefficient is the ratio of the difference between the maximum and minimum daily electricity consumption to the average daily electricity consumption. Based on the user's basic dataset and the number of days with electricity consumption characteristics, the average daily heat supply, temperature adjustment frequency, and heat loss coefficient within the user's electricity consumption characteristics are obtained. The temperature adjustment frequency is the number of times the user changes the daily temperature setting value, and the heat loss coefficient is the ratio of the difference between the heat supply and the actual heat absorbed indoors to the heat supply. Based on user demand response data and historical cogeneration data, and using the number of characteristic days of electricity consumption as a basis, the response participation rate, response intensity, and response delay time are obtained. The response participation rate is the ratio of the actual number of responses to the number of response invitations. The response intensity is the ratio of the average load reduction per response to the baseline load. The response delay time represents the average time from the issuance of the response invitation to the start of the response. The user's basic data is normalized using the average daily electricity consumption, peak-valley electricity consumption ratio, electricity fluctuation coefficient, average daily heat supply, temperature regulation frequency, heat loss coefficient, response participation rate, response intensity, and response delay time as feature dimensions to obtain the strategy feature vector corresponding to each user. The policy feature vectors corresponding to the same user are divided into the same dataset to obtain the policy feature vector dataset information; Based on the information in the strategy feature vector dataset, and through data clustering feature analysis, a data clustering matching algorithm is obtained. Based on the data clustering matching algorithm, data clustering is performed on the policy feature vector to obtain data clustering information; Based on data clustering information, users are classified to obtain user classification information.

3. The method for optimizing the scheduling of a combined heat and power system based on dynamic game theory according to claim 2, characterized in that, The step of obtaining the number of days of electricity consumption characteristics for each user based on the user base dataset specifically includes: Based on the user base dataset, obtain the historical electricity consumption data from each user base dataset. The historical electricity consumption data includes the user's daily peak and off-peak electricity consumption and the total daily electricity consumption. Based on the user's daily peak and off-peak electricity consumption, obtain the user's daily peak and off-peak electricity consumption information and the corresponding electricity consumption information for each peak and off-peak period; Based on the electricity consumption corresponding to any peak period of a user's daily peak and valley periods, the ratio of the electricity consumption during the valley period to the electricity consumption corresponding to the peak period of the user's daily peak and valley periods is used as the user's electricity consumption deviation coefficient. The ratio of the minimum to the maximum value of the user's daily peak and off-peak electricity consumption deviation coefficient is used as the user's electricity consumption stability difference coefficient for that day. Based on the electricity consumption stability difference coefficient and the user basic dataset, obtain the electricity consumption characteristic coefficient corresponding to each user; Adjust the electricity consumption characteristic coefficient until it reaches its maximum value, and obtain the number of electricity consumption characteristic days for each user; Specifically, the electricity consumption characteristic coefficient is: In the formula, Electricity consumption characteristic coefficient This represents the maximum value of the electricity consumption stability difference coefficient within h days. This represents the minimum value of the electricity consumption stability difference coefficient within h days, where h is the number of characteristic days for electricity consumption.

4. The method for optimizing the scheduling of a combined heat and power system based on dynamic game theory according to claim 3, characterized in that, The step of obtaining a data clustering matching algorithm based on the strategy feature vector dataset information and data clustering feature analysis specifically includes: Based on the information in the strategy feature vector dataset, and using the KS test, we determine whether each feature dimension in the strategy feature vector dataset conforms to a normal distribution, and use the feature dimensions that conform to a normal distribution as the benchmark feature dimensions. Based on the baseline feature dimension, the strategy feature vector corresponding to the baseline feature dimension is used as the baseline feature vector, and the strategy feature vector that is not the baseline feature vector is used as the secondary feature vector. Based on the baseline feature vectors, any two baseline feature vectors are used as the baseline feature group; The sum of the Euclidean distances between the baseline eigenvector and all secondary eigenvectors is used as the eigendivergence coefficient of the baseline eigenvector. The benchmark feature vector corresponding to the larger feature divergence coefficient in the benchmark feature group is taken as the first feature vector, and the benchmark feature vector corresponding to the smaller feature divergence coefficient is taken as the second feature vector. Based on the baseline feature set, obtain the Pearson correlation coefficient between two baseline feature vectors in the baseline feature set; The benchmark feature vectors in the benchmark feature group are filtered based on the Pearson correlation coefficient to obtain the calibration feature vectors; Wherein, if the absolute value of the Pearson correlation coefficient is less than 0.3, the baseline feature vector in the baseline feature group is a non-redundant feature; if the absolute value of the Pearson correlation coefficient is not less than 0.3, the baseline feature vector in the baseline feature group is a redundant feature, and the second feature vector in the baseline feature group is removed. The ratio of the number of calibrated feature vectors to the total number of policy feature vectors is used as the normal distribution coefficient; Based on the normal distribution coefficient, obtain the data clustering matching algorithm; If the normal distribution coefficient is greater than 0.8, the K-means algorithm is used as the data clustering matching algorithm; if the normal distribution coefficient is not greater than 0.8, the DBSCAN algorithm is used as the data clustering matching algorithm.

5. The method for optimal scheduling of a combined heat and power system based on dynamic game theory according to claim 4, characterized in that, The data clustering matching algorithm, which performs data clustering on the strategy feature vectors, specifically includes: Based on the data clustering matching algorithm, obtain the corresponding data clustering parameters; If the data clustering matching algorithm is the K-means algorithm, then the initial number of clusters is obtained based on the policy feature vector dataset information and the elbow method. Based on the data clustering requirements of the K-means algorithm, K-means iteration parameters are set, including the maximum number of iterations, the cluster center convergence threshold, and the initial cluster center selection method. If the data clustering matching algorithm is the DBSCAN algorithm, the neighborhood radius is obtained based on the elbow method according to the information of the policy feature vector dataset, and the minimum number of samples is set. Based on the data clustering parameters, the stability of the data clustering matching algorithm is verified. Based on stratified sampling, 70% of the data samples are extracted from the policy feature vector dataset as data samples and the clustering is repeated 5 times. The adjusted Rand index of each clustering result and the original result is calculated. Based on the adjusted Rand index, it is determined whether the data clustering matching algorithm meets the data clustering requirements. If the adjusted Rand index is greater than 0.8, the data clustering is considered stable; if the adjusted Rand index is not greater than 0.8, the data clustering is considered unstable, and the data clustering parameters are adjusted accordingly. Specifically, if the data clustering matching algorithm is a K-means algorithm, the number of iterations is increased or the initial center selection method is adjusted; if the data clustering matching algorithm is a K-means algorithm, the radius is decreased or the minimum number of samples is reduced. Based on the data clustering matching algorithm and data clustering parameters, the policy feature vectors are clustered to obtain data clustering information, which includes data cluster information, the central policy feature vector information corresponding to each data cluster, and the number of users within the cluster.

6. The method for optimal scheduling of a combined heat and power system based on dynamic game theory according to claim 5, characterized in that, The process of classifying users based on data clustering information and obtaining user classification information specifically includes: Based on the data clustering information, obtain the central strategy feature vector information corresponding to each data cluster; The Euclidean distance between the central policy feature vectors of any two data clusters is taken as the policy difference between the two data clusters. Based on data clustering information and strategy dissimilarity, construct the dissimilarity matrix D: ； In the formula, For the first The data cluster and the first The strategy variance of each data cluster; based on the variance matrix, obtain the mean and standard deviation of all elements in the variance matrix; Obtain the difference threshold based on the mean and standard deviation; Based on the user base dataset, the data interaction frequency corresponding to each data cluster is obtained. The data interaction frequency represents the number of times that the load adjustment direction of users in the same cluster is consistent with that of users in other clusters during the same period. Based on the data interaction frequency corresponding to each data cluster, 1 / 3 of the average data interaction frequency corresponding to the data cluster is used as the data interaction threshold. Based on the difference threshold and the data interaction threshold, obtain the strategy isolation index corresponding to each data cluster; Based on the preset policy isolation index threshold and the policy isolation index corresponding to each data cluster, determine whether the data cluster is an isolated data cluster and obtain user classification information; Specifically, the difference threshold is: In the formula, The difference threshold, The mean of the difference matrix, The standard deviation of the difference matrix; The isolation index of the strategy is specifically: In the formula, Let x be the policy isolation index for the x-th data cluster. Let x be the maximum policy difference between the x-th data cluster and other data clusters. This represents the data interaction frequency of the x-th data cluster. For data interaction threshold, Here are the weighting coefficients, where .

7. A dynamic game-based optimization scheduling system for a combined heat and power (CHP) system, used to implement the optimization scheduling method as described in any one of claims 1-6, characterized in that, include: The main control module is used to obtain corresponding data clustering parameters according to the data clustering matching algorithm, verify the stability of the data clustering matching algorithm according to the data clustering parameters, perform data clustering on the strategy feature vector according to the data clustering matching algorithm and the data clustering parameters to obtain data clustering information, classify users based on the data clustering information to obtain user classification information, train the existing model according to the user basic data corresponding to the main group of users to obtain the group demand prediction model, obtain the benchmark power supply prediction load and benchmark heating prediction load based on the benchmark power supply load and benchmark heating prediction load, take the benchmark power supply prediction load and the second power supply prediction load as the total power supply demand load, take the benchmark heating prediction load and the second heating prediction load as the total heating demand load, and schedule the cogeneration system according to the total power supply demand load and the total heating demand load. The information acquisition module is used to acquire information about the cogeneration system, including functional information and performance parameters of the cogeneration system. Based on the cogeneration system information, it acquires historical cogeneration data, including historical power supply data and historical heating data. It also acquires user information and basic user data, including historical electricity consumption data, historical heating data, and demand response data for each user. Finally, it acquires real-time user load data, including real-time electricity load data and real-time heating load data. The evaluation module is used to divide the user basic data corresponding to the same user into the same dataset based on the historical data of cogeneration and user basic data, obtain the user basic dataset, obtain the electricity consumption characteristic days corresponding to each user based on the user basic dataset, determine whether each feature dimension in the strategy feature vector dataset conforms to a normal distribution based on the KS test method based on the information of the strategy feature vector dataset, take the feature dimension that conforms to the normal distribution as the benchmark feature dimension, and obtain the data clustering matching algorithm based on the information of the strategy feature vector dataset and based on data clustering feature analysis. The display module interacts with the main control module and is used to output and display user classification information, real-time user load data, baseline power supply forecast load, baseline heating forecast load, second power supply forecast load, second heating forecast load, total power supply demand load, and total heating demand load.

8. The optimized scheduling system for a combined heat and power system based on dynamic game theory according to claim 7, characterized in that, The main control module specifically includes: The control unit is used to classify users based on data clustering information, obtain user classification information, train an existing model based on the user basic data corresponding to the main group of users, obtain a group demand prediction model, obtain a benchmark power supply prediction load and a benchmark heating prediction load based on the benchmark power supply load and the benchmark heating prediction load, take the benchmark power supply prediction load and the second power supply prediction load as the total power supply demand load, take the benchmark heating prediction load and the second heating prediction load as the total heating demand load, and schedule the cogeneration system according to the total power supply demand load and the total heating demand load. An information receiving unit, which interacts with the information acquisition module and the evaluation module, is used to receive data and transmit it to the data processing unit. The data processing unit is used to obtain corresponding data clustering parameters according to the data clustering matching algorithm, verify the stability of the data clustering matching algorithm according to the data clustering parameters, and perform data clustering on the strategy feature vector according to the data clustering matching algorithm and the data clustering parameters to obtain data clustering information.

9. The optimized scheduling system for a combined heat and power system based on dynamic game theory according to claim 7, characterized in that, The information acquisition module specifically includes: The first acquisition unit is used to acquire cogeneration system information, which includes cogeneration system function information and cogeneration system performance parameter information. Based on the cogeneration system information, the first acquisition unit is used to acquire cogeneration historical data, which includes historical power supply data and historical heating data. The second acquisition unit is used to acquire user information and user basic data. The user basic data includes historical electricity consumption data, historical heating data and user demand response data for each user. The second acquisition unit is used to acquire real-time load data of users. The real-time load data of users includes real-time electricity load data and real-time heating load data of users.

10. The optimized scheduling system for a combined heat and power system based on dynamic game theory according to claim 7, characterized in that, The evaluation module specifically includes: The first evaluation unit is used to divide the user basic data corresponding to the same user into the same dataset according to the historical data of cogeneration and user basic data, obtain the user basic dataset, and obtain the electricity consumption characteristic days corresponding to each user according to the user basic dataset. The second evaluation unit, based on the information in the strategy feature vector dataset and the KS test, determines whether each feature dimension in the strategy feature vector dataset conforms to a normal distribution. The feature dimension that conforms to a normal distribution is used as the benchmark feature dimension. Based on the information in the strategy feature vector dataset and the data clustering feature analysis, the data clustering matching algorithm is obtained.

Images