Integrated auxiliary frequency modulation intelligent control method and system for heat supply unit

Abstract

The present application belongs to the technical field of frequency modulation control, and particularly relates to a heat supply unit integrated auxiliary frequency modulation intelligent control method and system. The method obtains unit and heat supply subsystem operation parameters, combines a seasonal collaborative prediction coupling model with a built-in Bayesian model, obtains a seasonal prediction frequency modulation boundary space and accuracy, generates a seasonal load regulation distribution value based on the prediction result and frequency modulation evaluation rules using a multi-dimensional fuzzy decision optimization algorithm, obtains frequency modulation result deviation and decision success rate through real-time simulation and evaluation, and dynamically feeds back and optimizes the seasonal collaborative prediction coupling model or updates the control strategy list according to the evaluation result. The present application realizes intelligent collaborative control of the unit and the heat supply system, and significantly improves the response accuracy of the AGC load instruction and the system frequency modulation economy.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of frequency modulation control, and particularly relates to an integrated auxiliary frequency modulation intelligent control method and system for a heat supply unit. BACKGROUND

[0002] Under the background of energy structure transformation, ultra-supercritical heat supply units have become key equipment for realizing heat and power cogeneration and improving energy utilization efficiency. Such units not only need to meet power supply demand, but also need to undertake industrial heat supply tasks. The operation characteristics of such units are complex, involving multiple strongly coupled and large-lagged controlled objects such as unit coordination, steam temperature and heat supply system. The traditional control strategy usually controls the unit power generation and the heat supply system independently, which is difficult to effectively respond to frequent peak shaving and frequency modulation instructions of the power grid. In particular, when receiving AGC instructions, there is a lack of coordination between the unit coordination control system and the heat supply control system, and the heat storage capacity of the heat supply system cannot be utilized as a flexible frequency modulation resource, resulting in that the overall frequency modulation response speed, accuracy and economy of the unit do not reach the optimal level, which not only makes it difficult to meet the increasingly stringent grid evaluation rules, but also limits the flexibility and economic benefits of unit operation. SUMMARY

[0003] In view of the deficiencies of the prior art, the application provides an integrated auxiliary frequency modulation intelligent control method and system for a heat supply unit. The method obtains unit and heat supply subsystem operation parameters, combines a seasonal collaborative prediction coupling model with a built-in Bayesian model, obtains a seasonal prediction frequency modulation boundary space and accuracy, generates a seasonal load regulation distribution value based on the prediction result and a frequency modulation evaluation rule using a multi-dimensional fuzzy decision optimization algorithm, obtains frequency modulation result deviation and decision success rate through real-time simulation and evaluation, and dynamically feeds back and optimizes the seasonal collaborative prediction coupling model or updates the control strategy list according to the evaluation result. The application realizes intelligent collaborative control of the unit and the heat supply system, and significantly improves the response accuracy of the AGC load instruction and the system frequency modulation economy.

[0004] To achieve the above-mentioned purpose, the application provides the following technical scheme:

[0005] The integrated auxiliary frequency modulation intelligent control method for a heat supply unit comprises the following steps:

[0006] Obtaining unit and industrial heat supply subsystem operation parameters in an alternating current power distribution network, combining a seasonal collaborative prediction coupling model, obtaining a seasonal prediction frequency modulation boundary space and prediction accuracy;

[0007] The seasonal collaborative prediction coupling model is constructed by combining a coupling algorithm with a built-in Bayesian model according to seasonal mode characteristics of AGC load instructions at different time points in the alternating current power distribution network, unit and industrial heat supply subsystem interactive operation state information, and the product of prediction accuracy and seasonal frequency modulation success rate, and is used to predict a first frequency modulation boundary space under different seasonal mode characteristics.

[0008] Based on the in-season prediction frequency modulation boundary space and prediction accuracy, combined with the preset frequency modulation evaluation rules and multi-dimensional fuzzy decision optimization algorithm, the in-season load regulation distribution value of the AGC load instruction is obtained;

[0009] Based on the in-season load regulation distribution value, combined with the unit control subsystem and heating control subsystem respectively configured in the unit and industrial heating subsystem, real-time simulation is performed using a simulation algorithm, and the simulation process is evaluated in real time to obtain the current in-season frequency modulation result deviation value and the current in-season frequency modulation success rate.

[0010] Specifically, the auxiliary frequency modulation intelligent control method further comprises:

[0011] Obtain the operating state information of the unit and the industrial heating subsystem corresponding to the minimum in-season frequency modulation result deviation value in the simulation process, and obtain the in-season frequency modulation success rate of the next AGC load instruction by combining the preset in-season frequency modulation control list, the comprehensive evaluation algorithm and the current in-season frequency modulation success rate.

[0012] When the in-season frequency modulation result deviation value does not meet the corresponding AGC load instruction requirement, or the in-season frequency modulation success rate of the next AGC load instruction does not meet the preset in-season frequency modulation success rate threshold, the corresponding in-season load regulation distribution value and the current in-season frequency modulation success rate are fed back to the in-season collaborative prediction coupling model for adjustment.

[0013] If the in-season frequency modulation result deviation value meets the corresponding AGC load instruction requirement, and the in-season frequency modulation success rate of the next AGC load instruction meets the preset in-season frequency modulation success rate threshold, then the in-season load regulation distribution value of the AGC load instruction is mapped to the in-season frequency modulation control list to obtain the frequency modulation peak shaving control strategy of the current AGC load instruction.

[0014] Specifically, the construction and training process of the in-season collaborative prediction coupling model comprises:

[0015] Based on the AGC load instruction, unit operating parameters and industrial heating subsystem operating and interaction parameters at different time points, data cleaning and standardization preprocessing are performed to obtain a standardized time series data set;

[0016] Based on the standardized time series data set and combined with a hierarchical analysis algorithm, the in-season load regulation distribution value, constraint boundary, control parameter and decision frequency modulation success rate of each AGC load instruction analysis information and the unit and industrial heating subsystem at the corresponding time are obtained.

[0017] The initial frequency modulation instruction tree is constructed by taking each AGC load instruction analysis information as a root node, taking the constraint boundary corresponding to the unit and the industrial heating subsystem as a parallel first-level node, taking the seasonal load adjustment distribution value as a second-level node, taking the control parameter corresponding to the unit and the industrial heating subsystem as a parallel third-level node, and taking the decision frequency modulation success rate as a fourth-level node.

[0018] Specifically, the construction and training process of the seasonal collaborative prediction coupling model includes:

[0019] The first directed connection between different root nodes is constructed according to the time priority of the AGC load instruction, and the dynamic coupling connection between the parallel first-level nodes under the same AGC load instruction is constructed according to the constraint interaction information between the unit and the industrial heating subsystem and the coupling strength obtained by the coupling algorithm.

[0020] Based on the initial frequency modulation instruction tree at different time points, the first directed connection and the dynamic coupling connection, a directed frequency modulation instruction forest is constructed by a block chain algorithm and a random forest algorithm; the seasonal frequency modulation control list is associated with the directed frequency modulation instruction forest.

[0021] Specifically, the construction and training process of the seasonal collaborative prediction coupling model includes:

[0022] Based on the AGC load instruction analysis information at different time points, a multi-dimensional feature vector is extracted from the AGC load instruction analysis information at each time point by a time series feature extraction algorithm; the multi-dimensional feature vector at least includes: curve slope change state, fluctuation amplitude, duration, previous period typical load mean value and predicted next period typical load change trend;

[0023] Based on the extracted multi-dimensional feature vector, the multi-dimensional feature vector is processed by a principal component algorithm to obtain a principal clustering feature vector;

[0024] Based on the principal clustering feature vector and the adaptive optimal cluster number determination algorithm, the number K of seasonal mode clusters is determined; the adaptive optimal cluster number determination algorithm obtains the silhouette coefficient and Davies-Bouldin index corresponding to different K values through the curve slope change state, fluctuation amplitude and duration of the typical load curve in each mode, and selects the K value that makes the silhouette coefficient maximum and the Davies-Bouldin index minimum as the optimal cluster number;

[0025] Based on the directed frequency modulation instruction forest and the determined optimal cluster number, K-Means++ algorithm is used for clustering to obtain K seasonal mode clusters.

[0026] Specifically, the construction and training process of the seasonal collaborative prediction coupling model includes:

[0027] The AGC load instruction analysis information at different time points in each seasonal mode cluster and the operation information of the corresponding unit and industrial heat supply subsystem are combined with a time sequence algorithm to construct a unit regulation evaluation sub-model and a heat supply frequency modulation sub-model;

[0028] The unit load regulation capability is predicted based on the unit regulation evaluation sub-model combined with the operation data at continuous time points of the unit;

[0029] The constraint bidirectional mapping is obtained based on the current seasonal frequency modulation success rate and the coupling strength of the parallel first-level node under the same root node as the prior probability, combined with a Bayesian algorithm;

[0030] The predicted unit load regulation capability is mapped to the heat supply frequency modulation sub-model under the same root node for first coupling training constraint through the constraint bidirectional mapping.

[0031] Specifically, the construction and training process of the seasonal collaborative prediction coupling model also includes:

[0032] The heat supply frequency modulation upper limit and the heat supply rate upper limit of the industrial heat supply subsystem are predicted based on the heat supply frequency modulation sub-model combined with the heat supply operation data of the industrial heat supply subsystem at continuous time points of the unit, the heat supply cost change trend, and the heat supply operation data of the industrial heat supply subsystem at continuous time points of the unit, and the heat supply frequency modulation upper limit and the heat supply rate upper limit are mapped to the unit regulation evaluation sub-model under the same root node for second coupling training constraint through the constraint bidirectional mapping;

[0033] The seasonal collaborative prediction coupling model is constructed based on the unit regulation evaluation sub-model and the heat supply frequency modulation sub-model and the first coupling training constraint and the second coupling training constraint combined with an integrated training framework, and the training is performed with the maximum decision frequency modulation success rate and the prediction accuracy as the training target to obtain the trained seasonal collaborative prediction coupling model and the first frequency modulation boundary space under the corresponding seasonal mode.

[0034] Specifically, the specific steps of obtaining the seasonal load regulation allocation value of the AGC load instruction include:

[0035] Based on the seasonal prediction frequency modulation boundary space, the unit frequency modulation capability upper limit, the heat supply frequency modulation capability upper limit, and the system comprehensive regulation rate are extracted as the first group of input variables;

[0036] Based on the prediction accuracy, the prediction confidence is extracted as the second group of input variables; and based on the frequency modulation evaluation rule, the frequency modulation evaluation KPI weight is extracted as the third group of input variables;

[0037] A fuzzy rule base is established with the unit and heat network regulation cost, heat supply participation degree, system response speed, decision frequency modulation success rate, and evaluation standard reaching rate as optimization objectives;

[0038] obtaining a fuzzy set of output variables through a Mamdani type fuzzy reasoning method combined with the fuzzy rule base based on the first set of input variables, the second set of input variables, the third set of input variables; the output variables include a unit load distribution coefficient, a heating system frequency modulation participation factor, and a heat storage system release or absorption rate;

[0039] obtaining a load regulation distribution initial value through defuzzification calculation of the fuzzy set of output variables by using the gravity method; checking the load regulation distribution initial value in the constraint condition defined by the seasonal prediction frequency modulation boundary space, and triggering a fuzzy rule base re-reasoning mechanism if the constraint condition is violated until the output value meets all boundary constraints;

[0040] outputting the unit load distribution coefficient, the heating system frequency modulation participation factor, and the heat storage system release or absorption rate that pass the check as the final seasonal load regulation distribution value.

[0041] The intelligent control system for integrated auxiliary frequency modulation of a heating unit includes a boundary prediction module, a decision module, and a simulation module.

[0042] The boundary prediction module is configured to obtain operation parameters of a unit and an industrial heating subsystem in an alternating current power distribution network, and obtain a seasonal prediction frequency modulation boundary space and a prediction accuracy rate in combination with a seasonal collaborative prediction coupling model.

[0043] The seasonal collaborative prediction coupling model is constructed in combination with a coupling algorithm with a built-in Bayesian model based on seasonal mode characteristics of AGC load instructions at different time points in the alternating current power distribution network, interactive operation state information of the unit and the industrial heating subsystem, and the product of the prediction accuracy rate and a seasonal frequency modulation success rate, and is used to predict a first frequency modulation boundary space under different seasonal mode characteristics.

[0044] The decision module is configured to obtain a seasonal load regulation distribution value of the AGC load instruction based on the seasonal prediction frequency modulation boundary space and the prediction accuracy rate, in combination with a preset frequency modulation evaluation rule and a multi-dimensional fuzzy decision optimization algorithm.

[0045] The simulation module is configured to perform real-time simulation by using a simulation algorithm based on the seasonal load regulation distribution value in combination with a unit control subsystem and a heating control subsystem respectively configured in the unit and the industrial heating subsystem, and to perform real-time evaluation on the simulation process to obtain a current seasonal frequency modulation result deviation value and a current seasonal frequency modulation success rate.

[0046] Specifically, the system further includes an evaluation module and a discrimination module.

[0047] The evaluation module is configured to obtain operation state information of the unit and the industrial heating subsystem corresponding to the minimum seasonal frequency modulation result deviation value in the simulation process, and obtain a seasonal frequency modulation success rate of a next AGC load instruction by combining a preset seasonal frequency modulation control list, a comprehensive evaluation algorithm and a current seasonal frequency modulation success rate.

[0048] The discrimination module is configured to, based on a discrimination result that the seasonal frequency modulation result deviation value does not meet a requirement of the corresponding AGC load instruction or the seasonal frequency modulation success rate of the next AGC load instruction does not meet a preset seasonal frequency modulation success rate threshold, feed back the corresponding seasonal load adjustment distribution value and the current seasonal frequency modulation success rate to the seasonal collaborative prediction coupling model for adjustment.

[0049] The discrimination module is configured to, based on a discrimination result that the seasonal frequency modulation result deviation value does not meet a requirement of the corresponding AGC load instruction or the seasonal frequency modulation success rate of the next AGC load instruction does not meet a preset seasonal frequency modulation success rate threshold, feed back the corresponding seasonal load adjustment distribution value and the current seasonal frequency modulation success rate to the seasonal collaborative prediction coupling model for adjustment.

[0050] Compared with the prior art, the present application has the following beneficial effects:

[0051] The present application is aimed at the deficiencies of the prior art, and through the seasonal collaborative prediction coupling model, the seasonal mode of the AGC load instruction in different seasons can be accurately matched, the frequency modulation boundary space prediction is more in line with the seasonal operation law of the unit and the industrial heating subsystem, and the prediction pertinence is improved. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 The present application is aimed at the deficiencies of the prior art, and through the seasonal collaborative prediction coupling model, the seasonal mode of the AGC load instruction in different seasons can be accurately matched, the frequency modulation boundary space prediction is more in line with the seasonal operation law of the unit and the industrial heating subsystem, and the prediction pertinence is improved.

[0053] Figure 2 The present application is aimed at the deficiencies of the prior art, and through the seasonal collaborative prediction coupling model, the seasonal mode of the AGC load instruction in different seasons can be accurately matched, the frequency modulation boundary space prediction is more in line with the seasonal operation law of the unit and the industrial heating subsystem, and the prediction pertinence is improved. DETAILED DESCRIPTION

[0054] Embodiment 1:

[0055] Please refer toFigure 1 The application provides a heat supply unit integrated auxiliary frequency modulation intelligent control method, which comprises the following steps:

[0056] S1, obtain the operation parameters of the unit and the industrial heat supply subsystem in the alternating current power distribution network, combine the seasonal collaborative prediction coupling model, obtain the seasonal prediction frequency modulation boundary space and the prediction accuracy;

[0057] The seasonal collaborative prediction coupling model is obtained by combining the coupling algorithm with the built-in Bayesian model according to the seasonal mode characteristics of the AGC load instruction at different time points in the alternating current power distribution network, the interactive operation state information of the unit and the industrial heat supply subsystem, and the product of the prediction accuracy and the seasonal frequency modulation success rate, and is used to predict the first frequency modulation boundary space under different seasonal mode characteristics;

[0058] S2, based on the seasonal prediction frequency modulation boundary space and the prediction accuracy, combining the preset frequency modulation evaluation rules and the multi-dimensional fuzzy decision optimization algorithm, obtain the seasonal load regulation distribution value of the AGC load instruction;

[0059] S3, based on the seasonal load regulation distribution value, combining the unit control subsystem and the heat supply control subsystem respectively configured in the unit and the industrial heat supply subsystem, using the simulation algorithm for real-time simulation, and real-time evaluation of the simulation process, obtaining the current seasonal frequency modulation result deviation value and the current seasonal frequency modulation success rate;

[0060] It should be further pointed out that the current seasonal frequency modulation success rate in the embodiment is based on the process that the unit control subsystem and the heat supply control subsystem perform real-time simulation according to the seasonal load regulation distribution value under the current AGC load instruction, and the real-time evaluation of the simulation process is used to measure the probability that the result of the AGC load instruction after frequency modulation operation can meet the corresponding frequency modulation demand of the AGC load instruction;

[0061] S4, obtain the operation state information of the unit and the industrial heat supply subsystem corresponding to the minimum seasonal frequency modulation result deviation value in the simulation process, combine the preset seasonal frequency modulation control list, the comprehensive evaluation algorithm and the current seasonal frequency modulation success rate, and obtain the seasonal frequency modulation success rate of the next AGC load instruction;

[0062] It needs to be further explained that the success rate of the next AGC load instruction in the embodiment is based on the operating state information of the unit and the industrial heat supply subsystem corresponding to the minimum deviation value of the simulation process, combined with the preset seasonal frequency modulation control list, the comprehensive evaluation algorithm and the current seasonal frequency modulation success rate, to measure the probability that the result of the next AGC load instruction after frequency modulation operation can meet the frequency modulation demand of the next AGC load instruction under seasonal conditions; the operation includes: first, the probability of meeting the next instruction frequency modulation standard is grasped in advance, avoiding excessive frequency modulation deviation caused by blind execution, and ensuring the stability of the coordinated operation of the unit and the industrial heat supply subsystem; second, relying on the optimal operating state and the current success rate data, the prediction is more in line with the seasonal operating characteristics, and the seasonal adaptability of the frequency modulation strategy is improved; third, the quantitative probability guides the strategy adjustment, reduces the parameter trial and error cost in actual frequency modulation, ensures that the frequency modulation process meets the demand of the grid AGC instruction, and does not break the operating boundary of the unit and the heat supply system, realizes the two-way guarantee of frequency modulation reliability and heat supply stability;

[0063] S5, when the seasonal frequency modulation result deviation value does not meet the corresponding AGC load instruction demand, or the seasonal frequency modulation success rate of the next AGC load instruction does not meet the preset seasonal frequency modulation success rate threshold, the corresponding seasonal load regulation distribution value and the current seasonal frequency modulation success rate are fed back to the seasonal coordinated prediction coupling model for adjustment; if the seasonal frequency modulation result deviation value meets the corresponding AGC load instruction demand, and the seasonal frequency modulation success rate of the next AGC load instruction meets the preset seasonal frequency modulation success rate threshold, then the seasonal load regulation distribution value of the AGC load instruction is mapped to the seasonal frequency modulation control list, and the frequency modulation peak shaving control strategy of the current AGC load instruction is obtained.

[0064] In the integrated auxiliary frequency modulation scene of the heat supply unit, the seasonal instruction matches the AGC load instruction characteristics, the unit operating parameter constraints and the industrial heat supply subsystem operating law according to the climate conditions and the industrial heat supply demand changes in different seasons, so that the frequency modulation and peak shaving strategy formulation, model training, parameter setting and the actual operating characteristics of the current season are deeply adapted. The core is to realize the coordinated optimization of frequency modulation and heat supply system around the seasonal differentiated operating conditions, and to avoid using a unified standard to deal with the operating condition differences in different seasons.

[0065] The seasonal analysis is performed because there are seasonal differences in heat supply demand and unit constraints. First, in winter, the industrial heat supply load is high, and the heat supply subsystem needs to prioritize the stability of water temperature and flow. Due to the task of supplying heat extraction steam, the operating range of parameters such as main steam pressure and secondary reheat temperature is more stringent, and the frequency modulation capability boundary is narrowed. In the non-heating season, the heat supply demand is low, and the heat supply subsystem has high adjustment flexibility. The unit parameter adjustment margin increases, and the frequency modulation response space is wider. If the seasons are not distinguished, it will lead to a disconnection between the frequency modulation and peak shaving strategy and the actual system constraints. Second, in winter, due to the heating demand, the power grid load fluctuates more frequently and with a larger amplitude, the AGC instruction change rate is high, and the duration is unstable. In the non-heating season, the power grid load fluctuation is relatively smooth, and the AGC instruction characteristics are more regular. The unified model cannot accurately fit the instruction fluctuation law in different seasons, which will reduce the prediction accuracy of the frequency modulation boundary.

[0066] In the embodiment, the seasonal analysis has key technical value in the integrated auxiliary frequency modulation of heat supply units and can improve the system operation efficiency in multiple dimensions. In particular, it enables the seasonal collaborative prediction coupling model to be trained based on the seasonal exclusive AGC instruction characteristics, unit and heat supply parameter constraints, avoids the deviation of the unified model in fitting seasonal difference data, improves the prediction accuracy of the first frequency modulation boundary space, and provides a reliable constraint basis for subsequent load regulation distribution. In winter, by setting strict heat supply parameter constraints, the influence of frequency modulation operation on heat supply quality is prevented, and in the non-heating season, the unit frequency modulation response parameters are optimized to improve efficiency, realizing priority adaptation in seasonal working conditions and ensuring the bidirectional stability of heat supply and frequency modulation. The KPI weight of frequency modulation evaluation can also be adjusted according to seasonal differences, so that the seasonal load regulation distribution value output by the multi-dimensional fuzzy decision optimization algorithm is more in line with the current power grid evaluation focus and system constraints, reducing resource waste and parameter trial and error costs. At the same time, by dividing the seasonal mode cluster, the directed frequency modulation instruction forest and the seasonal frequency modulation control list accurately match the seasonal working conditions, improving the reliability of the current seasonal frequency modulation success rate prediction, avoiding low frequency modulation compliance rate or heat supply failure caused by blind execution, and enhancing the overall operation adaptability and reliability of the system.

[0067] It needs to be further explained that the construction and training process of the seasonal collaborative prediction coupling model in the embodiment includes:

[0068] S101, data cleaning and standardization preprocessing are performed based on the AGC load instruction, unit operating parameters and industrial heat supply subsystem operating and interaction parameters at different time points to obtain a standardized time series data set;

[0069] It needs to be further explained that the AGC load instruction in the embodiment includes an instruction value, a change rate, and a duration, which is used as a reference instruction for frequency modulation and peak regulation of the unit and the industrial heating subsystem, triggers the control strategy of the corresponding season, and clearly defines the load regulation target and distribution direction; the unit operation parameters include main steam pressure, main steam temperature, secondary reheat temperature, turbine valve position, turbine speed, boiler fuel quantity, and boiler feed water flow, which are used to monitor the unit operation state in real time, provide unit operation data support for the seasonal collaborative prediction coupling model, assist in adjusting the unit fuel supply, valve opening, and other actuator actions, and ensure the safe and stable operation of the unit during frequency modulation; the industrial heating subsystem operation and interaction parameters include heating and water supply pressure, heating and water supply flow, heating return water temperature, heating extraction steam pressure, heating extraction steam flow, heat storage tank pressure, and heat storage tank temperature, which are used to monitor the industrial heating quality and provide heating boundary conditions for load regulation distribution, and assist in adjusting the heating valve opening and heat storage tank throughput to realize stable heating and frequency modulation collaboration.

[0070] S102, based on the normalized time series data set combined with the hierarchical analysis algorithm, the AGC load instruction analysis information and the corresponding time unit and the industrial heating subsystem seasonal load regulation distribution value, constraint boundary, control parameter and decision frequency modulation success rate are obtained;

[0071] It needs to be further explained that the analysis process in the embodiment includes:

[0072] Based on the AGC load instruction data in the normalized time series data set, the instruction fluctuation period, peak interval, and change trend characteristics are extracted through a feature extraction algorithm to obtain AGC load instruction analysis information;

[0073] Based on the unit operation parameters and the industrial heating subsystem operation and interaction parameters in the normalized time series data set, the seasonal operation limit value and the safety deviation range of each parameter are calculated through a boundary threshold calculation algorithm to obtain the constraint boundary of the unit and the industrial heating subsystem at the corresponding time; it needs to be further explained that the unit operation parameters include but are not limited to main steam pressure, secondary reheat temperature, and turbine valve position; the operation and interaction parameters include but are not limited to heating and water supply pressure, flow, and heat storage tank pressure;

[0074] Based on the seasonal load regulation distribution value and the constraint boundary, a mapping relationship between the seasonal load regulation distribution value and the unit fuel quantity, valve opening, and heating valve opening, and heat storage throughput rate is established through a control parameter mapping algorithm to obtain the control parameters of the unit and the industrial heating subsystem at the corresponding time;

[0075] Based on the AGC load instruction analysis information, seasonal load regulation distribution value, constraint boundary, and control parameter, the decision frequency modulation success rate at the corresponding time is obtained through a Bayesian probability evaluation algorithm combined with historical collaborative frequency modulation success probability data.

[0076] S103, taking each AGC load instruction analysis information as a root node, taking the constraint boundary corresponding to the unit and the industrial heat supply subsystem as a parallel first-level node, taking the seasonal load adjustment allocation value as a second-level node, taking the control parameter corresponding to the unit and the industrial heat supply subsystem as a parallel third-level node, and taking the decision frequency modulation success rate as a fourth-level node, combining a tree-shaped database, and constructing an initial frequency modulation instruction tree;

[0077] S104, constructing a first directed connection between different root nodes according to the time priority of the AGC load instruction, and constructing a dynamic coupling connection between the parallel first-level nodes under the same AGC load instruction according to the constraint interaction information between the unit and the industrial heat supply subsystem and the coupling strength obtained by the coupling algorithm;

[0078] It needs to be further explained that the construction process of the dynamic coupling connection in the embodiment includes:

[0079] Based on the constraint interaction information of the unit and the industrial heat supply subsystem under the same AGC load instruction, the correlation degree between each interaction parameter is calculated through a grey correlation analysis algorithm to obtain an initial coupling coefficient set. The constraint interaction information includes but is not limited to the associated deviation value of the main steam pressure of the unit and the heat supply extraction steam pressure, the coupling data of the unit load regulation rate and the heat supply flow rate, and the interaction influence quantity of the heat storage tank temperature and the secondary reheat temperature.

[0080] Based on the initial coupling coefficient set, the weight factor of each interaction parameter is determined by an analytic hierarchy process, the initial coupling coefficient is weighted and summed with the corresponding weight factor to obtain a comprehensive coupling strength value.

[0081] Based on the comprehensive coupling strength value, the necessity of the connection is judged by a threshold comparison algorithm, and when the comprehensive coupling strength value is greater than or equal to the threshold value, a dynamic coupling connection is established between the parallel first-level nodes of the unit and the industrial heat supply subsystem.

[0082] Based on the dynamic coupling connection, the constraint boundary deviation value is taken as a correction variable, and the connection weight is dynamically updated in combination with the decision frequency modulation success rate, wherein the connection weight value = comprehensive coupling strength value x (1+ deviation correction coefficient), and a dynamic coupling connection with real-time weight attribute is obtained. The relative deviation between the actual value and the target value of the constraint boundary is taken as the basic input, and a minimum-maximum normalization algorithm is used to map it to the [0, 1] interval to obtain a normalized deviation quantity. At the same time, the decision frequency modulation success rate is converted into a reliability coefficient between 0 and 1 through a Sigmoid function. Finally, the deviation correction coefficient is the product of the normalized deviation quantity and the reliability coefficient, multiplied by a preset proportional gain coefficient Kp. The proportional gain coefficient Kp in the embodiment is determined by measuring the system response speed and the running stability.

[0083] S105, based on the initial frequency modulation instruction tree of different time points, combining the first directed connection, dynamically coupled connection, through the blockchain algorithm and random forest algorithm, a directed frequency modulation instruction forest is constructed; the seasonal frequency modulation control list is associated with the directed frequency modulation instruction forest;

[0084] It needs to be further explained that the construction process of the seasonal frequency modulation control list in the embodiment includes:

[0085] Based on the directed frequency modulation instruction forest, through the forest feature extraction algorithm, the AGC load instruction analysis information of the root node of each initial frequency modulation instruction tree, the constraint boundary of the parallel first level node, the comprehensive coupling strength value of the dynamic coupling connection, the time priority label of the first directed connection and the seasonal mode cluster label of each node are extracted, and a directed frequency modulation instruction forest key feature set is obtained;

[0086] Based on the directed frequency modulation instruction forest key feature set and the seasonal frequency modulation control list, wherein the seasonal frequency modulation control list contains control entries divided by seasons, each entry contains a seasonal mode label, a frequency modulation control parameter benchmark value and a constraint boundary threshold value, through a feature-entry matching algorithm, the seasonal mode cluster label in the directed frequency modulation instruction forest key feature set is matched with the seasonal mode label of the control list entry for consistency, and a preliminary association mapping table is established;

[0087] Based on the preliminary association mapping table, through a connection feature weighting algorithm, the comprehensive coupling strength value of the dynamic coupling connection in the directed frequency modulation instruction forest is mapped to the cooperative adjustment coefficient of the corresponding entry of the control list, and the time priority label of the first directed connection is mapped to the parameter adjustment weight of the corresponding entry of the control list, to generate an association mapping table with weight attribute;

[0088] Based on the association mapping table with weight attribute, through an association verification algorithm, historical frequency modulation data is called to compare the historical frequency modulation execution result corresponding to the node in the directed frequency modulation instruction forest with the preset control effect of the associated entry of the control list, calculate the association error, and select the effective association pair with the association error less than the preset threshold value;

[0089] Based on the real-time collected unit and industrial heating subsystem operation data, through an association dynamic updating algorithm, the frequency modulation control parameter benchmark value, the constraint boundary threshold value of the control list entry in the effective association pair and the weight factor corresponding to the forest feature are iteratively corrected, to maintain the real-time association effectiveness of the seasonal frequency modulation control list and the directed frequency modulation instruction forest.

[0090] It needs to be further explained that, in the embodiment, based on the initial frequency modulation instruction tree, the first directed connection and the dynamic coupling connection at different time points, the key time sequence data can be prevented from being tampered with through the decentralized storage and hash encryption characteristics of the blockchain algorithm, and a traceable data link is formed, ensuring the integrity and authenticity of the data used when constructing the instruction forest later, providing a data trust basis for the reliable generation of the directed frequency modulation instruction forest, and facilitating the subsequent tracing of the original data source of each node and connection in the instruction forest, assisting in troubleshooting and data auditing. Based on the trusted data guaranteed by the blockchain algorithm, the multi-dimensional features such as the initial frequency modulation instruction tree, the first directed connection and the dynamic coupling connection can be mined through the multi-decision tree ensemble learning feature of the random forest algorithm, the influence weight of different features on the instruction forest structure can be automatically learned, and then multiple instruction decision trees that are correlated and have strong generalization ability are generated to form the directed frequency modulation instruction forest. At the same time, the anti-overfitting feature of the random forest algorithm can avoid the distortion of the instruction forest structure caused by a single instruction tree or local data bias, ensuring that the generated directed frequency modulation instruction forest can accurately reflect the correlation logic between the AGC load instruction and the unit and the heating system, and providing structural support for the subsequent precise correlation with the seasonal frequency modulation control list. The blockchain algorithm provides trusted input data for the random forest algorithm, and the random forest algorithm realizes effective structure construction based on the trusted data. The combination of the two can ensure that the directed frequency modulation instruction forest has data-level credibility, structure-level rationality and practicality, and ultimately guarantees the subsequent correlation between the directed frequency modulation instruction forest and the seasonal frequency modulation control list, realizing the precise guidance of the AGC load instruction frequency modulation strategy.

[0091] S106, based on the AGC load instruction analysis information at different time points, a multi-dimensional feature vector is extracted from the AGC load instruction analysis information at each time point through a time sequence feature extraction algorithm; the multi-dimensional feature vector at least includes: curve slope change state, fluctuation amplitude, duration, previous period typical load mean value and predicted next period typical load change trend;

[0092] S107, based on the extracted multi-dimensional feature vector, a principal component algorithm is used to perform dimension reduction processing on the multi-dimensional feature vector to obtain a principal clustering feature vector;

[0093] S108, based on the principal clustering feature vector, the number K of seasonal mode clusters is determined by using an adaptive optimal cluster number determination algorithm; the adaptive optimal cluster number determination algorithm obtains the silhouette coefficient and Davies-Bouldin index corresponding to different preset K values through the curve slope change state, fluctuation amplitude and duration of the typical load curve in each mode, and selects the K value that makes the silhouette coefficient maximum and the Davies-Bouldin index minimum as the optimal cluster number;

[0094] S109, based on the determined optimal number of clusters, clustering is performed through a K-Means++ algorithm to obtain K seasonal pattern clusters;

[0095] S110, based on the AGC load instruction analysis information at different time points in each seasonal pattern cluster and the operation information of the corresponding unit and industrial heating subsystem, a time series algorithm is used to construct a unit regulation evaluation sub-model and a heating frequency modulation sub-model;

[0096] It should be further explained that in the embodiment, the input of the unit regulation evaluation sub-model is the AGC load instruction analysis information at different time points in the corresponding seasonal pattern cluster and the operation data of the unit at consecutive time points, including main steam pressure, main steam temperature, secondary reheat temperature, turbine governor valve position, boiler fuel quantity, boiler feed water flow, unit actual load, and when constructing, a time series algorithm is used, such as ARIMA algorithm or LSTM neural network algorithm, through trend analysis, periodicity mining and correlation modeling on the input time series data, the output is the predicted unit load regulation capability, including but not limited to unit maximum load regulation amplitude, load regulation rate upper limit, load regulation response delay time, and unit operation stability parameters under different load regulation amounts; the input of the heating frequency modulation sub-model is the AGC load instruction analysis information at different time points in the corresponding seasonal pattern cluster and the operation information of the industrial heating subsystem, including but not limited to heating water supply pressure, heating water supply flow, heating return water temperature, heating extraction steam pressure, heating extraction steam flow, heat storage tank pressure, heat storage tank temperature, and when constructing, a time series algorithm is also used, preferably a GRU neural network algorithm or a SARIMA algorithm, through seasonal characteristic analysis, heating parameter and AGC instruction correlation modeling and heating load fluctuation law mining on the input time series data, the output is the predicted heating subsystem frequency modulation related parameters, including heating frequency modulation load regulation range, frequency modulation response speed under the premise of heating parameter stability, heating extraction steam quantity regulation limit, and heat throughput capacity of heat storage tank auxiliary frequency modulation.

[0097] S111, based on the unit regulation evaluation sub-model and the operation data of the unit at consecutive time points, the unit load regulation capability is predicted and obtained;

[0098] S112, based on the current seasonal frequency modulation success rate and the coupling strength of the parallel primary nodes under the same root node as the prior probability, a constraint bidirectional mapping is obtained through a Bayesian algorithm;

[0099] The embodiment is based on the current seasonality frequency modulation success rate and the coupling strength of the parallel first-level node under the same root node, and a two-way association bridge between the unit regulation evaluation sub-model and the heating frequency modulation sub-model is constructed by a Bayesian algorithm. On the one hand, the unit load regulation capability predicted by the unit regulation evaluation sub-model is mapped to the heating frequency modulation sub-model, which provides the first coupling training constraint conforming to the unit bearing capacity, so as to avoid that the heating frequency modulation operation exceeds the unit regulation limit. On the other hand, the heating frequency modulation upper limit and the heating rate upper limit predicted by the heating frequency modulation sub-model are mapped to the unit regulation evaluation sub-model, which provides the second coupling training constraint conforming to the heating safety and stability requirement, so as to prevent the unit load regulation from affecting the heating quality. The effect is to realize the collaborative coupling training of the two sub-models based on the same AGC load instruction root node, to ensure that the prediction results of both sides conform to the unit operation capability boundary and meet the heating system constraint requirement, and to improve the collaboration and accuracy of the subsequent frequency modulation strategy.

[0100] It needs to be further explained that the construction process of the constraint two-way mapping includes:

[0101] Step 1, based on the current season, the seasonality frequency modulation success rate of the next AGC load instruction predicted by the previous AGC load instruction is obtained, and the seasonality constraint interaction information of the parallel first-level node corresponding to the unit regulation evaluation sub-model and the heating frequency modulation sub-model under the same root node is obtained. The related deviation value of the unit main steam pressure and the heating extraction steam pressure in the heating season, and the interaction data of the unit exhaust steam temperature and the heating pipe network loss in the non-heating season, the coupling strength of the parallel first-level node in the season is calculated by the grey correlation analysis algorithm, and the seasonality prior probability required by the Bayesian algorithm is obtained, including the current seasonality frequency modulation success rate and the seasonality coupling strength.

[0102] Step 2, based on the seasonality prior probability, a seasonality prior distribution model is constructed by a Bayesian algorithm, wherein the prior distribution parameters are calibrated based on the unit-heating collaborative frequency modulation historical data of the same period in the past three years in the current season, wherein the data from November to the following March is used in the heating season, and the data from April to October is used in the non-heating season. The seasonality posterior probability is calculated, and the seasonality association basis between the unit regulation evaluation sub-model and the heating frequency modulation sub-model is established.

[0103] Step 3, based on the current season exclusive unit load regulation capability output by the unit regulation evaluation sub-model, wherein the maximum unit load regulation range and the load regulation rate upper limit considering the extraction constraint in the heating season, and the unit load response delay time and the operation stability parameters considering the exhaust constraint in the non-heating season, combined with the seasonality posterior probability, the seasonality load regulation capability is converted into the seasonality constraint threshold of the heating frequency modulation sub-model by a linear mapping algorithm, wherein the heating extraction amount is constrained to be not more than 95% of the unit extraction upper limit in the heating season, and the heating return water temperature deviation is constrained to be not more than ±2℃ in the non-heating season, to generate the first coupling training constraint.

[0104] Step 4, based on the current season-specific heating frequency modulation upper limit output by the heating frequency modulation sub-model and the heating rate upper limit, wherein the heating season is the heat storage priority heating frequency modulation load adjustment range, the heat storage tank heat throughput capacity upper limit, and the non-heating season is the energy consumption priority heating extraction steam quantity adjustment limit and the frequency response speed upper limit, combined with the season posterior probability, the season-specific heating frequency modulation upper limit and the heating rate upper limit are converted into the season-specific constraint conditions of the unit regulation evaluation sub-model through a nonlinear mapping algorithm, specifically: the heating season constraint unit fuel quantity increase / decrease amplitude does not affect the heating extraction steam pressure, and the non-heating season constraint unit valve position regulation rate does not cause the heating water flow fluctuation to exceed ±5%; the second coupling training constraint is generated;

[0105] Step 5, based on the unit-heating collaborative frequency modulation historical compliance data of the current season, including the heating season taking historical heating compliance rate ≥98% data and the non-heating season taking historical frequency modulation compliance rate ≥95% data, the deviation value of the first coupling training constraint, the second coupling training constraint and the historical season constraint is checked through a deviation checking algorithm, if the deviation value exceeds the season deviation threshold of the current season, the season posterior probability is iteratively updated based on the Bayesian algorithm, the constraint value is adjusted again by executing steps 3-4, until the deviation value meets the season deviation threshold requirement, and finally the constraint bidirectional mapping result is obtained.

[0106] S113, mapping the predicted unit load regulation capability to the heating frequency modulation sub-model under the same root node as the first coupling training constraint through the constraint bidirectional mapping;

[0107] S114, based on the heating frequency modulation sub-model, combining the heating operation data and the heating cost change trend of the industrial heating subsystem at the continuous time points of the unit, predicting the heating frequency modulation upper limit and the heating rate upper limit of the industrial heating subsystem, and mapping the heating frequency modulation upper limit and the heating rate upper limit to the unit regulation evaluation sub-model under the same root node as the second coupling training constraint through the constraint bidirectional mapping;

[0108] S115, based on the unit regulation evaluation sub-model, the heating frequency modulation sub-model, the first coupling training constraint and the second coupling training constraint, combining the integrated training framework, constructing the season-specific collaborative prediction coupling model, and taking the maximum decision frequency modulation success rate and prediction accuracy as the training target to train, obtaining the trained season-specific collaborative prediction coupling model and the corresponding first frequency modulation boundary space under the season-specific mode.

[0109] It needs to be further explained that the construction process of the second coupling training constraint in the embodiment includes:

[0110] Based on the heating operation data of the industrial heating subsystem at consecutive time points, through a seasonal data preprocessing algorithm, the data at consecutive time points is aligned by a sliding window, abnormal data is removed by a seasonal outlier identification algorithm, missing data is completed by a seasonal interpolation model, and a seasonal standardized heating operation time series data set is obtained; the heating operation data includes heating water supply pressure, heating water supply flow, heating return water temperature, heating extraction steam pressure, heating extraction steam flow, heat storage tank pressure, heat storage tank temperature, pipe network pressure loss along the line, and user end heat load feedback value; the seasonal outlier identification algorithm removes abnormal data based on the standard deviation threshold of operating parameters in different seasons, specifically removing abnormal data caused by pipe network freezing in winter and pipe network leakage in summer. The sliding window seasonal alignment uses a fixed length window to segment and align the data at consecutive time points, and the window length is determined by autocorrelation analysis of historical data. For example, a 7-day cycle window is used in winter and a 14-day cycle window is used in summer.

[0111] Based on the seasonal standardized heating operation time series data set, the seasonal fluctuation item, long-term trend item and random disturbance item of each operating parameter are separated by an STL-seasonal trend decomposition algorithm, the parameter correlation degree is calculated by a mutual information entropy-seasonal correlation algorithm, and a heating operation seasonal multi-dimensional feature vector is extracted; the parameter correlation degree calculated by the mutual information entropy-seasonal correlation algorithm is specifically the correlation degree of winter heat storage tank temperature-extraction steam flow and the correlation degree of summer water supply flow-pipe network loss. The mutual information entropy-seasonal correlation algorithm specifically calculates the mutual information value of winter heat storage tank temperature-extraction steam flow and summer water supply flow-pipe network loss, and selects strong correlation features with mutual information value > 0.6.

[0112] Based on the heating cost change trend data, the importance of each cost item to heating adjustment is evaluated by using a gradient boosting tree-seasonal weight algorithm, and the weight is corrected by a seasonal cost fluctuation coefficient to obtain a seasonal heating cost dynamic influence weight vector; the heating cost change trend data includes seasonal heating fuel unit cost, circulating pump energy consumption cost, heat exchanger energy consumption cost, heat storage tank insulation energy consumption cost, seasonal equipment maintenance cost, and heat load deficiency compensation cost; the seasonal cost fluctuation coefficient is specifically the winter fuel cost fluctuation coefficient and the summer energy consumption cost fluctuation coefficient; the gradient boosting tree-seasonal weight algorithm is realized based on XGBoost, the seasonal heating cost data is used as a label, the model is trained to evaluate the importance of fuel cost and energy consumption cost, the tree depth is set to 3-5, and the learning rate is 0.1. The seasonal cost fluctuation coefficient is determined based on the analytic hierarchy process, a cost influence-seasonal adaptation judgment matrix is constructed, the winter fuel cost fluctuation coefficient and the summer energy consumption cost fluctuation coefficient are calculated, and the consistency ratio CR of the judgment matrix is less than 0.1.

[0113] Based on the heating operation seasonal multi-dimensional feature vector and the seasonal heating cost dynamic influence weight vector, through a double-input seasonal fusion LSTM algorithm, the heating operation seasonal multi-dimensional feature vector is input into a seasonal feature extraction layer, the seasonal heating cost dynamic influence weight vector is input into a seasonal cost adaptation layer, multi-head attention mechanism is used to realize operation parameter-cost seasonal interaction fusion, a seasonal gate unit is introduced to strengthen the prediction bias, and a seasonal frequency adjustment boundary prediction model of the heating subsystem is constructed; the prediction bias strengthened by the seasonal gate unit is specifically a winter heat storage priority prediction bias and a summer energy consumption priority prediction bias.

[0114] Based on the seasonal frequency modulation boundary prediction model, combined with the seasonal pattern cluster matching algorithm, the corresponding seasonal pattern cluster of the current season in the directed frequency modulation instruction forest is retrieved, the seasonal statistical characteristics of the historical heating frequency modulation upper limit and the historical heating rate upper limit in the seasonal pattern cluster are extracted, and the upper limit constraint seasonal calculation algorithm is used to obtain the preliminary heating frequency modulation upper limit and the heating rate upper limit, that is, the heating frequency modulation upper limit=seasonal heat load carrying limit value×(1-division seasonal cost overrun warning coefficient), and the heating rate upper limit=seasonal parameter response speed threshold×(1+heat storage tank seasonal adjustment elasticity coefficient). The retrieval and extraction content of the seasonal pattern cluster matching algorithm is to retrieve the corresponding seasonal pattern cluster of the current season in the directed frequency modulation instruction forest, and extract the seasonal statistical characteristics of the historical heating frequency modulation upper limit and the historical heating rate upper limit in the cluster.It needs to be further explained that the acquisition process of the seasonal thermal load bearing limit value and the seasonal parameter response speed threshold value in the embodiment is as follows: based on the AGC load instruction data in the historical N and more complete operation cycles, the real-time operation data of the unit and the industrial heating subsystem (including heating load, equipment operation state, and environmental temperature), a directed frequency modulation instruction forest is constructed, through the seasonal mode cluster matching algorithm, according to the actual temperature interval and the heating demand level of the current season, the seasonal mode cluster matching the label in the directed frequency modulation instruction forest is searched, and the industrial heating demand level of the corresponding season is searched; wherein each node in the directed frequency modulation instruction forest is labeled with seasonal characteristics; secondly, M historical heating frequency modulation upper limits and historical heating rate upper limit data of the same seasonal cycle are extracted from the seasonal mode cluster matching the label, after removing the abnormal values by using the Grubbs criterion, the remaining valid data are respectively subjected to maximum value screening, and the maximum value of the historical heating frequency modulation upper limit obtained by screening is the seasonal thermal load bearing limit value, and the maximum value of the historical heating rate upper limit is the seasonal parameter response speed threshold value; the acquisition process of the seasonal cost overrun early warning coefficient is as follows: the heating cost detail data (including fuel procurement cost, equipment operation and maintenance cost, and energy consumption cost) of the same season in the history and the cost budget data of the corresponding season are collected, the cost overrun rate in each historical cycle is calculated, that is, the difference between the actual cost and the budget cost is divided by the budget cost, and a probability distribution model of the overrun rate is constructed by combining the Bayesian algorithm; at the same time, the heating cost budget target (determined based on the current fuel price and operation plan) and the cost redundancy (difference between the budget amount and the minimum necessary cost) of the current season are combined, the overrun rate distribution and the cost redundancy are mapped by a risk assessment algorithm, and the risk coefficient obtained by the mapping is the seasonal cost overrun early warning coefficient; the acquisition process of the thermal storage tank seasonal adjustment elasticity coefficient is as follows: the operation data of the thermal storage tank in the same season in the history (including charging and discharging power, tank temperature change value, heating load fluctuation in the corresponding period, and response delay time) are extracted, the relationship curve between the charging and discharging adjustment amount of the thermal storage tank and the heating load fluctuation is fitted by a linear regression algorithm, the slope mean value of the curve is calculated, and the slope mean value is corrected in combination with the actual thermal insulation performance test data (obtained by a heat loss rate experiment) of the thermal storage tank in the current season, and the corrected value is the thermal storage tank seasonal adjustment elasticity coefficient.

[0115] Based on the real-time collected current seasonal heating subsystem operation deviation data, that is, the deviation value of the current operation parameter and the seasonal mode cluster reference parameter, through the seasonal dynamic correction algorithm, the seasonal deviation feedback coefficient is used to iteratively adjust the preliminary heating frequency modulation upper limit and the heating rate upper limit, and the final seasonal heating frequency modulation upper limit and the seasonal heating rate upper limit of the heating subsystem are obtained; the seasonal deviation feedback coefficient is characterized in that the winter deviation feedback coefficient is higher than the summer deviation feedback coefficient.

[0116] Based on the final seasonal heating frequency regulation upper limit and the seasonal heating rate upper limit, by constraining the bidirectional mapping-seasonal adaptation algorithm, the upper limit value is associated with the seasonal operation boundary of the unit regulation evaluation sub-model under the same node to generate a second coupling training constraint; the seasonal operation boundary of the unit regulation evaluation sub-model is specifically the winter unit steam extraction constraint and the summer unit steam exhaust constraint.

[0117] The process is pre-processed by time series data, and the data is aligned by the seasonal cycle window determined by the autocorrelation function. Combined with the precise removal of winter and summer specific abnormal data by Isolation Forest, and the interpolation to complete the missing values, the integrity and accuracy of the seasonal data set are effectively improved, eliminating data quality interference for subsequent modeling; seasonal feature decomposition separates features such as seasonal fluctuations by STL, and combines mutual information value to filter winter and summer strong correlation parameters, reducing redundant features, making the seasonal multi-dimensional feature vector more consistent with seasonal working condition rules, and providing accurate input for model learning; seasonal cost quantification evaluates the importance of cost items by XGBoost, and combines AHP to determine the winter and summer cost fluctuation coefficient, so that the cost weight adapts to seasonal differences, avoids economic imbalance caused by uniform weight, and considers both frequency regulation and economy. The dual-input seasonal fusion LSTM is processed by dual-branch processing, multi-head attention interaction and seasonal gating, so that the prediction model is consistent with the seasonal core target, and the frequency regulation boundary prediction accuracy is improved; seasonal pattern matching retrieves historical cluster data of the same season, and combines constraint formulas to calculate the preliminary upper limit, ensuring that the initial value meets the seasonal load and cost constraints; seasonal dynamic correction iteratively adjusts the upper limit to adapt to real-time working conditions with a higher deviation feedback coefficient in winter; finally, the seasonal boundary of the unit is associated by bidirectional mapping constraint to form a second coupling training constraint. The overall process from data to model, from prediction to correction to coupling, adapts to seasonal characteristics layer by layer, not only improves the prediction accuracy of the heating frequency regulation upper limit and the rate upper limit, but also considers the economic cost and unit-heating collaborative stability, avoiding the deviation of frequency regulation or heating risk caused by seasonal working condition disconnection.

[0118] It needs to be further explained that the process of obtaining the seasonal load regulation distribution value of the AGC load instruction in the embodiment includes:

[0119] Based on the in-season prediction frequency modulation boundary space, the upper limit of unit frequency modulation capacity, the upper limit of heating frequency modulation capacity, and the system comprehensive adjustment rate are extracted as the first group of input variables; it needs to be further explained that the specific variables corresponding to the upper limit of unit frequency modulation capacity include the variables directly related to the upper limit in the unit adjustment evaluation sub-model output, including: the maximum load adjustment range of the unit, that is, the maximum difference of load adjustment that can be achieved by the unit within the safe operation boundary, the upper limit of load adjustment rate, that is, the maximum load adjustment amount that can be completed by the unit per unit time, the upper limit of unit operation stability related parameters under different load adjustment amounts, such as the maximum allowable fluctuation threshold of main steam pressure when the unit performs frequency modulation operation, the maximum adjustment limit of turbine valve position, the maximum increase and decrease range of boiler fuel quantity, the maximum adjustment range of boiler feed water flow, the maximum continuous adjustment time length that can be maintained in the continuous frequency modulation process of the unit, that is, the longest frequency modulation operation time of the unit without overloading or parameter exceeding; the specific variables corresponding to the upper limit of heating frequency modulation capacity include the variables directly related to the upper limit in the heating frequency modulation sub-model output, including: the upper limit value of the heating frequency modulation load adjustment range, that is, the maximum load adjustment amount that can be involved in frequency modulation by the heating subsystem under the premise of ensuring heating quality, the adjustment limit of heating extraction steam quantity, that is, the maximum increase and decrease range of extraction steam quantity that can be achieved by the heating extraction steam system, the upper limit of heat throughput capacity of the heat storage tank auxiliary frequency modulation, that is, the maximum heat value that can be released or stored by the heat storage tank per unit time, the upper limit of frequency modulation response speed under the premise of heating parameter stability, that is, the maximum frequency modulation response rate that can be achieved by the heating subsystem without causing the heating water supply pressure and return water temperature to exceed the standard, the maximum allowable fluctuation upper limit of heating water supply pressure, the maximum allowable deviation upper limit of heating return water temperature, that is, the boundary upper limit of water supply pressure and return water temperature that needs to be maintained by the heating subsystem when performing frequency modulation operation, and cannot normally participate in frequency modulation if exceeded.

[0120] extract a prediction confidence as a second group of input variables based on the prediction accuracy; and extract a frequency modulation examination KPI weight as a third group of input variables based on the frequency modulation examination rule; the frequency modulation examination KPI weight is determined based on a power grid peak regulation and frequency modulation examination rule, and is used to quantify the importance of different frequency modulation examination core indicators in the overall examination system; wherein the frequency modulation examination core KPI includes: a frequency modulation response speed KPI, which measures the speed of response of the unit and the industrial heat supply subsystem to the AGC load instruction, such as whether the examination requirement of response delay ≤ preset delay threshold is met, a frequency modulation regulation accuracy KPI, which measures the deviation degree of the actual load regulation amount and the AGC instruction regulation amount, such as whether the examination requirement of regulation accuracy ≥ preset standard is met, a frequency modulation capacity KPI, which measures the maximum frequency modulation regulation amount that can be provided by the unit and the heat supply subsystem, such as whether the demand standard of the power grid for the frequency modulation capacity is reached, an examination compliance rate KPI, which measures the proportion of the number of times that the frequency modulation operation meets the power grid examination requirement in the total frequency modulation times in a unit time, and a heat supply cooperation stability KPI, which measures the operating parameter stability of the industrial heat supply subsystem in the frequency modulation process, such as whether the heat supply and water supply pressure and the backwater temperature deviation are within the allowed range; the KPI weight is the proportion value allocated to each of the above core KPIs according to the different seasonal scheduling requirements of the power grid, such as focusing on heat supply stability in the heating season to increase the weight of the heat supply cooperation stability KPI, and focusing on frequency modulation efficiency in the non-heating season to increase the weight of the frequency modulation response speed KPI, so as to clearly define the priority of different examination indicators in the multi-dimensional fuzzy decision optimization algorithm and ensure that the seasonal load regulation distribution value matches the current examination focus of the power grid.

[0121] establish a fuzzy rule base with the unit and heat network regulation cost, heat supply participation degree, system response speed, decision frequency modulation success rate and examination compliance rate as optimization objectives;

[0122] based on the first group of input variables, the second group of input variables and the third group of input variables, obtain a fuzzy set of output variables through forward reasoning of the Mamdani-type fuzzy reasoning method combined with the fuzzy rule base; the output variables include a unit load distribution coefficient, a heat supply system frequency modulation participation factor and a heat storage system release or absorption rate;

[0123] It should be further explained that the process of forward reasoning of the Mamdani-type fuzzy reasoning method combined with the fuzzy rule base in the embodiment includes:

[0124] The first group of input variables, the second group of input variables, and the third group of input variables are respectively subjected to fuzzy processing: a fuzzy language variable is defined for each input variable, such as high, medium, and low for the upper limit of the unit frequency modulation capability, high, medium, and low for the prediction confidence, large, medium, and small for the frequency modulation KPI weight, and a corresponding membership function, such as a triangular membership function or a trapezoidal membership function, is configured for each fuzzy language variable, the exact value of each input variable is substituted into the membership function, and the membership value of each input variable to each fuzzy language variable is calculated;

[0125] Each rule is constructed in the form of "if the fuzzy state of the first group of input variables, and the fuzzy state of the second group of input variables, and the fuzzy state of the third group of input variables, then the fuzzy state of the unit and heat network regulation cost, and the fuzzy state of the heat supply participation degree, and the fuzzy state of the system response speed, and the fuzzy state of the decision frequency modulation success rate, and the fuzzy state of the examination compliance rate", the membership values of the current input variables are judged one by one whether they meet the fuzzy conditions of the antecedents of each rule, and all active rules that meet the conditions are selected;

[0126] For each active rule, the minimum value of the membership values of the input variables in the rule antecedent is determined as the trigger strength of the rule by using the minimum operation, and then the fuzzy set of the rule consequent output variables is cut by using the trigger strength, to obtain the output fuzzy subset corresponding to each active rule; finally, the fuzzy subsets of the same output variable, such as the unit and heat network regulation cost, the heat supply participation degree, etc., output by all active rules are merged by using the maximum operation, and the fuzzy set of each output variable is finally obtained.

[0127] The barycentric method is used to perform defuzzification calculation on the fuzzy set of the output variable to obtain an initial value of load regulation distribution; the initial value of load regulation distribution is substituted into the constraint conditions defined by the seasonal prediction frequency modulation boundary space for verification, and if the constraint conditions are violated, a fuzzy rule base re-reasoning mechanism is triggered until the output value meets all boundary constraints;

[0128] The unit load distribution coefficient, the heat supply system frequency modulation participation factor, and the heat storage system release or absorption rate that pass the verification are output as the final seasonal load regulation distribution value.

[0129] It should be further explained that the process of obtaining the seasonal frequency modulation success rate of the next AGC load instruction in the embodiment includes:

[0130] Based on the real-time collected seasonal frequency modulation result deviation data in the simulation process, the unit operating state information and the industrial heat supply subsystem operating state information corresponding to the minimum seasonal frequency modulation result deviation value are obtained by using a deviation extreme value screening algorithm;

[0131] Based on the obtained unit and industrial heating subsystem operation state information, a directed frequency modulation instruction forest node matching algorithm is used to retrieve a target initial frequency modulation instruction tree node corresponding to the operation state information in the directed frequency modulation instruction forest, wherein the target initial frequency modulation instruction tree node contains root node AGC load instruction analysis information, first-level node constraint boundaries, and node-associated seasonal mode cluster tags;

[0132] Based on the target initial frequency modulation instruction tree node and the dynamic coupling connection of the node in the directed frequency modulation instruction forest, a seasonal frequency modulation control list-forest feature mapping algorithm is used to match a seasonal frequency modulation control parameter benchmark value corresponding to the dynamic coupling strength and seasonal mode cluster tag of the node in a preset seasonal frequency modulation control list.

[0133] Based on the seasonal frequency modulation control parameter benchmark value, the current seasonal frequency modulation success rate, the first directed connection of the target initial frequency modulation instruction tree node, and the operation state information of the corresponding unit and industrial heating subsystem, a multi-dimensional weighted comprehensive evaluation algorithm is used to calculate a basic seasonal decision frequency modulation probability of the next AGC load instruction.

[0134] Based on the basic seasonal decision frequency modulation probability, the historical frequency modulation success rate data of the seasonal mode cluster to which the target initial frequency modulation instruction tree node belongs in the directed frequency modulation instruction forest, and the feature similarity between the next AGC load instruction output by the seasonal cooperative prediction coupling model and the cluster, a historical data correction algorithm is used to dynamically adjust the basic seasonal decision frequency modulation probability to obtain a seasonal frequency modulation success rate of the next AGC load instruction.

[0135] The embodiment constructs a seasonal collaborative prediction coupling model of a fusion Bayesian model, combines time series algorithm modeling and bidirectional constraint mapping of the unit and the industrial heating subsystem, can accurately predict the frequency modulation boundary space based on the seasonal pattern characteristics, avoids breaking the unit operation limit and the heating quality boundary during frequency modulation operation, guarantees the seasonal adaptability and safety of the basic prediction, relies on the tamper-proof storage and traceability of the initial frequency modulation instruction tree, dynamic coupling connection and other data based on the blockchain algorithm, combines the multi-decision tree integration and anti-overfitting ability of the random forest algorithm to construct a directed frequency modulation instruction forest, which can not only ensure data reliability to support subsequent correlation matching, but also accurately reflect the correlation logic of the AGC instruction, the unit and the industrial heating subsystem, and provide a reliable correlation carrier for the seasonal frequency modulation control list; the AGC instruction characteristics and system constraint boundaries are extracted through a hierarchical analysis algorithm, the fuzzy processing of multiple input variables, the activation rule screening and the barycentric method are introduced into the Mamdani type fuzzy reasoning, and the KPI weight of the frequency modulation examination is introduced to realize multi-objective optimization, so that the seasonal load regulation distribution value considering the adjustment cost, the heating participation degree and the examination standard can be output, and the violation can be avoided through boundary verification; based on the deviation extreme value screening of real-time simulation, the optimal operating state is obtained, combined with the node matching of the directed frequency modulation instruction forest and the parameter mapping of the seasonal frequency modulation control list, the current seasonal frequency modulation success rate of the next AGC instruction is predicted through multi-dimensional evaluation and historical data correction, which can avoid deviation caused by blind execution in advance and reduce trial and error cost; finally, the seasonal collaborative prediction coupling model is optimized through a feedback adjustment mechanism when the deviation or success rate is not up to standard, and the control strategy is output when it is up to standard, forming a closed loop control, and realizing the accuracy of the frequency modulation response of the heating unit, the stability of the system operation and the adaptability of the grid examination demand, achieving the collaborative guarantee of frequency modulation and heating.

[0136] Embodiment 2:

[0137] Please refer to Figure 2 , the application provides another embodiment: a heating unit integrated auxiliary frequency modulation intelligent control system, comprising: a boundary prediction module, a decision module, a simulation module, an evaluation module and a discrimination module;

[0138] The boundary prediction module obtains the operating parameters of the unit and the industrial heating subsystem in the alternating current power distribution network, combines the seasonal collaborative prediction coupling model, obtains the seasonal prediction frequency modulation boundary space and the prediction accuracy, and obtains the seasonal prediction frequency modulation boundary space and the prediction accuracy.

[0139] The seasonal collaborative prediction coupling model is constructed according to the seasonal pattern characteristics of the AGC load instruction at different time points in the alternating current power distribution network, the interactive operating state information of the unit and the industrial heating subsystem, and the product of the prediction accuracy and the seasonal frequency modulation success rate, and a coupling algorithm with a built-in Bayesian model, and is used to predict the first frequency modulation boundary space under different seasonal pattern characteristics.

[0140] The decision module obtains an in-season load regulation distribution value of the AGC load instruction based on the in-season prediction frequency modulation boundary space and the prediction accuracy, in combination with a preset frequency modulation evaluation rule and a multi-dimensional fuzzy decision optimization algorithm.

[0141] The simulation module adopts a simulation algorithm to perform real-time simulation and analog, and to perform real-time evaluation on the simulation process, based on the in-season load regulation distribution value, in combination with a unit control subsystem and a heating control subsystem respectively configured in the unit and the industrial heating subsystem, to obtain a current in-season frequency modulation result deviation value and a current in-season frequency modulation success rate.

[0142] The evaluation module obtains operation state information of the unit and the industrial heating subsystem corresponding to a minimum in-season frequency modulation result deviation value in the simulation process, in combination with a preset in-season frequency modulation control list, a comprehensive evaluation algorithm and the current in-season frequency modulation success rate, to obtain an in-season frequency modulation success rate of a next AGC load instruction.

[0143] The discrimination module feeds back the corresponding in-season load regulation distribution value and the current in-season frequency modulation success rate to the in-season collaborative prediction coupling model for adjustment, based on a discrimination result that the in-season frequency modulation result deviation value does not meet a requirement of the corresponding AGC load instruction, or the in-season frequency modulation success rate of the next AGC load instruction does not meet a preset in-season frequency modulation success rate threshold.

[0144] The AGC load instruction in-season load regulation distribution value is mapped to the in-season frequency modulation control list to obtain a frequency modulation peak regulation control strategy of the current AGC load instruction, based on a discrimination result that the in-season frequency modulation result deviation value meets the requirement of the corresponding AGC load instruction, and the in-season frequency modulation success rate of the next AGC load instruction meets the preset in-season frequency modulation success rate threshold.

[0145] The embodiments of the present application are described above in combination with the drawings, but the present application is not limited to the above specific embodiments, and the above specific embodiments are only illustrative but not restrictive, and the person of ordinary skill in the art can make changes, modifications, replacements and variations to the above embodiments without departing from the purpose of the present application and the scope protected by the claims, and these are all within the protection of the present application.

Claims

1. A method for integrated auxiliary frequency modulation intelligent control of a heat supply unit, characterized in that, include: The operating parameters of generating units and industrial heating subsystems in the AC distribution network are obtained, and combined with the seasonal collaborative prediction coupling model, the seasonal prediction frequency regulation boundary space and prediction accuracy are obtained. The seasonal collaborative prediction coupling model is constructed based on the seasonal mode characteristics of AGC load commands at different time points in the AC distribution network, the interactive operation status information of the unit and the industrial heating subsystem, and the product of prediction accuracy and seasonal frequency regulation success rate, combined with a coupling algorithm with a built-in Bayesian model. It is used to predict the first frequency regulation boundary space under different seasonal mode characteristics. Based on the seasonal predicted frequency regulation boundary space and prediction accuracy, combined with the preset frequency regulation assessment rules and multi-dimensional fuzzy decision optimization algorithm, the seasonal load adjustment allocation value of AGC load instruction is obtained. Based on the seasonal load adjustment allocation value, and combined with the unit control subsystem and heating control subsystem respectively configured in the unit and industrial heating subsystem, a simulation algorithm is used to perform real-time simulation, and the simulation process is evaluated in real time to obtain the deviation value of the current seasonal frequency regulation result and the success rate of the current seasonal frequency regulation. The system obtains the operating status information of the unit and industrial heating subsystem corresponding to the minimum deviation value of seasonal frequency regulation results during the simulation process. Combined with the preset seasonal frequency regulation control list, comprehensive evaluation algorithm and current seasonal frequency regulation success rate, the system obtains the seasonal frequency regulation success rate of the next AGC load command. When the deviation value of the seasonal frequency regulation result does not meet the requirements of the corresponding AGC load instruction, or the success rate of seasonal frequency regulation of the next AGC load instruction does not meet the preset seasonal frequency regulation success rate threshold, the corresponding seasonal load adjustment allocation value and the current seasonal frequency regulation success rate are fed back to the seasonal collaborative prediction coupling model for adjustment. If the deviation value of the seasonal frequency regulation result meets the requirements of the corresponding AGC load instruction, and the success rate of seasonal frequency regulation of the next AGC load instruction meets the preset seasonal frequency regulation success rate threshold, then the seasonal load adjustment allocation value of the AGC load instruction is mapped to the seasonal frequency regulation control list to obtain the frequency regulation and peak shaving control strategy of the current AGC load instruction. The construction and training process of the seasonal collaborative prediction coupling model includes: Data cleaning and standardization preprocessing were performed based on AGC load commands, unit operating parameters, and industrial heating subsystem operating and interaction parameters at different time points to obtain a standardized time-series dataset. Based on a normalized time-series dataset and a hierarchical parsing algorithm, the parsing information of each AGC load instruction and the seasonal load adjustment allocation value, constraint boundary, control parameters and decision frequency regulation success rate of the unit and industrial heating subsystem at the corresponding time are obtained. Using the parsed information of each AGC load instruction as the root node, the constraint boundaries corresponding to the unit and the industrial heating subsystem as parallel first-level nodes, the seasonal load adjustment allocation value as the second-level node, the control parameters corresponding to the unit and the industrial heating subsystem as parallel third-level nodes, and the decision frequency regulation success rate as the fourth-level node, combined with the tree database, an initial frequency regulation instruction tree is constructed. The first directed connection between different root nodes is constructed according to the time priority of the AGC load instruction, and the dynamic coupling connection between parallel first-level nodes under the same AGC load instruction is constructed according to the coupling strength obtained by the coupling algorithm combined with the constraint interaction information between the unit and the industrial heat supply subsystem; The directed frequency modulation instruction forest is constructed by the block chain algorithm and the random forest algorithm based on the initial frequency modulation instruction tree at different time points, the first directed connection and the dynamic coupling connection; the seasonal frequency modulation control list is associated with the directed frequency modulation instruction forest; The specific steps for obtaining the seasonal load regulation distribution value of the AGC load instruction include: Based on the seasonal predicted frequency modulation boundary space, the upper limit of the unit frequency modulation capacity, the upper limit of the heat supply frequency modulation capacity and the system comprehensive regulation rate are extracted as the first group of input variables; Based on the prediction accuracy, the prediction confidence is extracted as the second group of input variables; based on the frequency modulation evaluation rule, the frequency modulation evaluation KPI weight is extracted as the third group of input variables; A fuzzy rule base is established with the unit and the heat network regulation cost, the heat supply participation degree, the system response speed, the decision frequency modulation success rate and the evaluation standard reaching rate as the optimization objectives; Based on the first group of input variables, the second group of input variables and the third group of input variables, the fuzzy set of output variables is obtained by the Mamdani type fuzzy reasoning method combined with the fuzzy rule base; the output variables include the unit load distribution coefficient, the heat supply system frequency modulation participation factor and the heat storage system release or absorption rate; The fuzzy set of output variables is de-fuzzied by the barycenter method to obtain the initial value of load regulation distribution; the initial value of load regulation distribution is substituted into the constraint conditions defined by the seasonal predicted frequency modulation boundary space for verification, and if the constraint conditions are violated, the fuzzy rule base re-reasoning mechanism is triggered until the output value meets all boundary constraints; The unit load distribution coefficient, the heat supply system frequency modulation participation factor and the heat storage system release or absorption rate that pass the verification are output as the final seasonal load regulation distribution value.

2. The integrated auxiliary frequency modulation intelligent control method for a heating unit according to claim 1, wherein, The construction and training process of the seasonal collaborative prediction coupling model further includes: Based on the AGC load instruction analysis information at different time points, a multi-dimensional feature vector is extracted from the AGC load instruction analysis information at each time point by a time series feature extraction algorithm; the multi-dimensional feature vector at least includes: curve slope change state, fluctuation amplitude, duration, previous period typical load mean value and predicted next period typical load change trend; Based on the extracted multi-dimensional feature vector, the multi-dimensional feature vector is processed by dimension reduction through a principal component algorithm to obtain a principal clustering feature vector; Based on the principal clustering feature vector, the number K of seasonal mode clusters is determined by an adaptive optimal cluster number determination algorithm; the adaptive optimal cluster number determination algorithm obtains the silhouette coefficient and Davies-Bouldin index corresponding to different K values through the curve slope change state, fluctuation amplitude and duration of the typical load curve in each mode, and selects the K value that makes the silhouette coefficient maximum and the Davies-Bouldin index minimum as the optimal cluster number. Based on the determined optimal cluster number and the directed frequency modulation instruction forest, clustering is performed through a K-Means++ algorithm to obtain K seasonal pattern clusters.

3. The integrated auxiliary frequency modulation intelligent control method for a heating unit according to claim 2, wherein, The construction and training process of the seasonal collaborative prediction coupling model further includes: Based on the AGC load instruction analysis information at different time points in each seasonal pattern cluster and the corresponding operation information of the unit and the industrial heating subsystem, a time series algorithm is used to construct a unit regulation evaluation sub-model and a heating frequency modulation sub-model. Based on the unit regulation evaluation sub-model and the operation data at consecutive time points of the unit, the unit load regulation capability is predicted. Based on the current seasonal frequency modulation success rate and the coupling strength of the parallel first-level node under the same root node as the prior probability, a Bayesian algorithm is used to obtain a constraint bidirectional mapping. The predicted unit load regulation capability is mapped to the heating frequency modulation sub-model under the same root node for first coupling training constraint through the constraint bidirectional mapping.

4. The integrated auxiliary frequency modulation intelligent control method for a heating unit according to claim 3, wherein, The construction and training process of the seasonal collaborative prediction coupling model further includes: Based on the heating frequency modulation sub-model, the heating operation data of the industrial heating subsystem at consecutive time points of the unit, the heating cost trend, the predicted upper limit of the heating frequency modulation of the industrial heating subsystem, and the upper limit of the heating rate are mapped to the unit regulation evaluation sub-model under the same root node for second coupling training constraint through the constraint bidirectional mapping. Based on the unit regulation evaluation sub-model, the heating frequency modulation sub-model, the first coupling training constraint, and the second coupling training constraint, an integrated training framework is used to construct a seasonal collaborative prediction coupling model, and the maximum decision frequency modulation success rate and prediction accuracy are used as training targets for training to obtain a trained seasonal collaborative prediction coupling model and a first frequency modulation boundary space under the corresponding seasonal pattern.

5. The integrated auxiliary frequency modulation intelligent control system for heating unit, which is used for realizing the integrated auxiliary frequency modulation intelligent control method for heating unit in any one of claims 1-4, characterized in that, It includes: a boundary prediction module, a decision module, and a simulation module. The boundary prediction module is configured to obtain unit and industrial heating subsystem operation parameters in an alternating power distribution network, combine a seasonal collaborative prediction coupling model, and obtain a seasonal prediction frequency modulation boundary space and a prediction accuracy. The seasonal collaborative prediction coupling model is constructed based on the seasonal pattern characteristics of AGC load instructions at different time points in the alternating power distribution network, the interactive operation state information of the unit and the industrial heating subsystem, and the product of the prediction accuracy and the seasonal frequency modulation success rate, and is constructed by combining a coupling algorithm with a built-in Bayesian model, and is used to predict the first frequency modulation boundary space under different seasonal pattern characteristics. The decision module is configured to combine a preset frequency modulation evaluation rule and a multi-dimensional fuzzy decision optimization algorithm based on the seasonal prediction frequency modulation boundary space and the prediction accuracy to obtain a seasonal load regulation allocation value of the AGC load instruction. The simulation module is configured to combine a unit control subsystem and a heating control subsystem respectively configured in the unit and the industrial heating subsystem, perform real-time simulation using a simulation algorithm, and perform real-time evaluation on the simulation process to obtain a current seasonal frequency modulation result deviation value and a current seasonal frequency modulation success rate.

6. The integrated auxiliary frequency modulation intelligent control system for heating units of claim 5, wherein, It also includes an evaluation module and a discrimination module. The evaluation module is configured to obtain operation state information of the unit and the industrial heating subsystem corresponding to a minimum value of a seasonal frequency modulation result deviation value in a simulation process, and obtain a seasonal frequency modulation success rate of a next AGC load instruction by combining a preset seasonal frequency modulation control list, a comprehensive evaluation algorithm, and a current seasonal frequency modulation success rate. The discrimination module is configured to, based on a discrimination result that the seasonal frequency modulation result deviation value does not meet a requirement of the corresponding AGC load instruction or the seasonal frequency modulation success rate of the next AGC load instruction does not meet a preset seasonal frequency modulation success rate threshold, feed back the corresponding seasonal load regulation distribution value and the current seasonal frequency modulation success rate to the seasonal cooperative prediction coupling model for adjustment. Based on a discrimination result that the seasonal frequency modulation result deviation value meets the requirement of the corresponding AGC load instruction and the seasonal frequency modulation success rate of the next AGC load instruction meets the preset seasonal frequency modulation success rate threshold, the seasonal load regulation distribution value of the AGC load instruction is mapped to the seasonal frequency modulation control list to obtain a frequency modulation and peak regulation control strategy of the current AGC load instruction.

Images