A smart control method and system for geothermal coupled solar heating stations

Abstract

This invention discloses an intelligent control method and system for a geothermal coupled solar heating station, belonging to the field of soil source heat pump operation and management technology. After fault diagnosis based on current operating parameters, an integrated algorithm prediction model is constructed using machine learning algorithms based on historical operating parameters. The system load for the next time step is output based on current meteorological conditions. A decision is made based on the prediction results and current operating parameters and transmitted to the load regulation module. After review by the load regulation module, the regulation decision is transmitted to the DCS system. By proactively adjusting operating conditions, the heating efficiency of the entire system is improved, while the impact of solar grid connection on grid stability is reduced, thus enhancing the overall system stability.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of soil source heat pump operation management, and particularly relates to an intelligent control method and system for a geothermal coupling solar heat supply station. BACKGROUND

[0002] Efficient and clean use of energy has been the goal pursued by countries. In recent years, clean energy that can be developed and utilized in a sustainable manner has developed rapidly, while fossil energy on Earth is gradually being depleted, and the development and utilization of new energy is imperative.

[0003] Solar energy and geothermal energy have the characteristics of continuous and stable energy supply, efficient recycling, and renewability, and have a wide range of applications in the heating, power generation and other industries. However, the renewable energy industry has always been characterized by extensive development and low utilization efficiency. With the comprehensive improvement of new technologies such as big data and artificial intelligence in the industry, the change of the past simple and extensive utilization of renewable energy is imminent, and the demand for energy utilization refinement and intelligentization is becoming more and more urgent.

[0004] In order to reduce carbon emissions and improve energy utilization efficiency, renewable energy is used for heating while generating electricity, but the intermittency and changes in user-side heat load have a great impact on the safety and efficiency of the heating system, so energy storage technology must be used for the heating unit to improve the load matching capability. The change of heating load and the mismatch of load will seriously affect the efficiency and safety of unit operation, so it is urgent to improve the response capability of the heating system to load changes. SUMMARY

[0005] In order to solve the defects in the prior art, the purpose of the present application is to provide an intelligent control method and system for a geothermal coupling solar heat supply station, which can predict the operating parameters of the energy system in advance and make corresponding adjustments, significantly improving the efficiency of the entire system.

[0006] The present application is achieved by the following technical solutions:

[0007] An intelligent control method for a geothermal coupling solar heat supply station, comprising:

[0008] S1: Monitor the operating parameters of the geothermal coupling solar heat supply station according to a preset monitoring period, and perform fault diagnosis on the obtained operating parameters; if the operating parameters are abnormal, compare the operating parameters with historical data, and determine a solution to the fault according to the comparison result; if the operating parameters are normal, go to S2;

[0009] S2: combine the historical normal operation parameters of the geothermal coupled solar heat supply station with the corresponding historical meteorological conditions as a data set, divide the data set into a training set and a validation set, and build an integrated algorithm prediction model; use the training set to train the integrated algorithm prediction model based on the integrated prediction algorithm of the nearest neighbor node algorithm, the random forest algorithm and the reinforcement learning algorithm, and use the validation set to optimize the integrated algorithm prediction model to predict the user end demand and the photovoltaic system power generation capacity;

[0010] S3: according to the normal operation parameters obtained in S1 and the prediction results of S2, the running conditions of each part of the geothermal coupled solar heat supply station are decided by calculating the checked flow, the cold and heat load and the power generation capacity;

[0011] S4: according to the decision results obtained in S3, the running parameters after decision are calculated, the running parameters after decision are compared with the fault data, after confirming that the power supply system and the heat supply system of the geothermal coupled solar heat supply station can normally operate, the adjustment signal is transmitted to the DCS system of the geothermal coupled solar heat supply station; if the power supply system and the heat supply system of the geothermal coupled solar heat supply station cannot normally operate after adjustment, the original working condition is maintained and the error adjustment data is uploaded.

[0012] Preferably, in S1, the operation parameters include selected time points, corresponding time point instantaneous temperatures, daily maximum air temperatures, daily minimum air temperatures, daily sunshine hours, daily air humidity, daily average wind speed and dates;

[0013] The operation parameters are normalized before fault diagnosis:

[0014]

[0015] In the formula, x i,j represents the jth parameter in the ith group of data; x' i,j represents the normalized parameter of x i,j ; min(x ,j ) represents the minimum value of the jth parameter in each group, and max(x ,j ) represents the maximum value of the jth parameter in each group.

[0016] Preferably, in S2, the training method of the prediction model comprises:

[0017] S2.1: according to the historical normal operation parameters of the geothermal coupled solar heat supply station, the corresponding historical meteorological conditions are imported, and the historical normal operation parameters and the historical meteorological conditions are normalized;

[0018] S2.2: use a correlation analysis method to analyze the main factors affecting the heat load, the cold load and the photovoltaic system power generation capacity, eliminate low-impact data, and use the remaining data as a data set;

[0019] S2.3: dividing the data set into a training set and a validation set using a cross-validation method;

[0020] S2.4: constructing an integrated algorithm prediction model according to the training set and the validation set, training a nearest neighbor algorithm prediction model, a random forest algorithm prediction model and an adaptive boosting algorithm prediction model respectively, and outputting the final prediction result by weighting the prediction results;

[0021] S2.5: the integrated algorithm prediction model updates the corresponding weight according to the prediction results of the nearest neighbor algorithm prediction model, the random forest algorithm prediction model and the adaptive boosting algorithm prediction model and the deviation of the actual operation parameters of the geothermal coupled solar heat supply station; when the prediction results of the integrated algorithm prediction model and the operation parameters deviate continuously for multiple times within a certain time, the integrated algorithm prediction model constructed in S2.4 is retrained;

[0022] S2.6: repeating steps S2.3-S2.5 to obtain a user-side heat demand prediction model, a user-side cold demand prediction model and a photovoltaic system power generation prediction model respectively.

[0023] Further preferably, S2.2 specifically comprises:

[0024] S2.2.1: according to the preset time step, using the grey correlation method to select the user heat load, the user cold load and the power generation of the photovoltaic array at the starting time point of each time step as the mother sequence; the subsequence selection time point, the corresponding time point instantaneous temperature, the daily maximum temperature, the daily minimum temperature, the daily sunshine time, the daily air humidity, the daily average wind speed and the date;

[0025] S2.2.2: according to the negative correlation between temperature, sunshine time and heat load, multiply the reciprocal operator when analyzing the correlation of heat load:

[0026] X i 'D1=[x i ' ,1 d1,x i ' ,2 d2,...,x i ' ,j d j ]

[0027]

[0028] Wherein, X i ' represents the i-th subsequence; x i ' ,j represents the j-th parameter in the i-th group of data; D1 is the reciprocal operator;

[0029] S2.2.3: Obtain the grey correlation degrees of each influencing factor on the user-side heat demand, the user-side cold demand and the power generation of the photovoltaic array by the grey correlation method respectively, remove the low-correlation-degree factors except the time points from the influencing factors; and take the processed data as the data set R, denoted as:

[0030]

[0031] Further preferably, S2.3 specifically comprises:

[0032] S2.3.1: Divide the data set R processed in step S2.2 using the cross-validation method, select 5-fold cross-validation, divide the data set R into 5 parts, of which 4 parts are used as the training set and 1 part is used as the validation set;

[0033] S2.3.2: Repeat step S2.3.1 for 5 times, each time selecting a different training set, and a total of 20 training sets and 5 validation sets are obtained.

[0034] Further preferably, S2.4 specifically comprises:

[0035] S2.4.1: Establish a nearest neighbor algorithm prediction model, set the hyperparameter k, select Manhattan distance as the distance measure, select mean squared error and mean absolute error as the loss function, select Gaussian function as the weighting based on the distance proximity, based on the training set and the validation set obtained in step S2.3, minimize the loss function as the target, use KD tree to accelerate the training, and obtain the nearest neighbor algorithm prediction model; denoted as:

[0036]

[0037]

[0038] wherein y i is the true value of the i-th parameter, is the predicted value of the i-th parameter, and n is the number of dimensions;

[0039]

[0040] wherein a, b and c are parameters, and x is the distance between a certain point in the space and the true value;

[0041] S2.4.2: Establish a random forest regression prediction model, initially select N tree feature trees, set the maximum depth of each feature tree to 50, and initially select the feature value of each feature tree as m try ; select the data set R obtained in step S2.2, use the bagging method to sample with replacement, generate N treeA training subset is used, and the unsampled out-of-bag data is used as the validation set; the CART method is used to divide the feature tree, and m factors are randomly selected from j influencing factors. try One influencing factor (m) try ≤j) is used as the splitting feature value of the current node of the feature tree. The mean squared error and mean absolute error minimization criteria are selected as the splitting criteria of the feature tree. The prediction result is the mean of the output of each tree. Iterative calculation is performed to optimize the number of feature trees and the feature values ​​of the decision tree with the goal of minimizing the error, until the error is less than the set threshold, and the random forest prediction model is obtained.

[0042] S2.4.3: Establish an adaptive boosting algorithm prediction model. Select the training set obtained in step S2.3, choose n training samples from it, and assign initial weights w to the training samples. 1i =1 / n,i=1,2,3,…,n, use W(1)=(w 11 ,w 12 ,w 13 ,…,w 1n The initial weights of the samples are represented by ), and D represents the number of learners; iterations are performed with d = 1, 2, 3, ..., D, to train the d-th weak learner H. d When (x), use the weak learner H d (x) Predict the regression error ε of the validation set output d Calculate the maximum error E of the samples on the training set. d and relative error ε di Calculate the weight α of the weak learner in the final learner. d According to weight α d Update the sample weights w(d+1); after training for D rounds, obtain D groups of weak learners H. d (x), the reinforcement learner h(x) is obtained by combining the weak learners according to their weights, and the number of weak learners D is optimized using the validation set data; denoted as:

[0043]

[0044]

[0045]

[0046]

[0047] W(d+1)=(w d+1,1 ,w d+1,2 ,w d+1,3 ,…,w d+1,n )

[0048]

[0049]

[0050]

[0051] wherein y i is the true value of the i-th parameter.

[0052] Further preferably, S2.5 specifically comprises:

[0053] S2.5.1: optimizing the integrated algorithm prediction model; inputting the validation set data obtained in step S2.3, using the trained nearest neighbor algorithm prediction model, random forest regression prediction model and adaptive boosting algorithm prediction model to obtain outputs respectively, taking the mean square error of the integrated algorithm model prediction result and the true value as the loss function, and using historical data to optimize the weighting proportion of each part until the loss function J θ is less than a set threshold:

[0054] D = w1d1 + w2d2 + w3d3

[0055]

[0056]

[0057] wherein D is the output result of the integrated algorithm prediction model; d i ,w i are the prediction results and weights of the nearest neighbor algorithm, random forest algorithm and adaptive boosting algorithm respectively; w' i is the updated weight; d' is the true value; and k is a parameter for controlling the weighting update;

[0058] S2.5.2: the integrated algorithm prediction model uploads each prediction result to the cloud database, and when the prediction result error or fault decision in a period of time exceeds the threshold, the cloud database sends a signal to remind the maintenance personnel to update the learning prediction module, and the prediction model is retrained by repeating the above steps S2.1-S2.5.1.

[0059] The application discloses an intelligent control system for a geothermal and solar energy heating station.

[0060] The operation checking module monitors the operation parameters of the geothermal and solar energy heating station according to a preset monitoring period, and performs fault diagnosis on the obtained operation parameters; if the operation parameters are abnormal, the operation parameters are uploaded to the cloud database; if the operation parameters are normal, the operation parameters are transmitted to the real-time decision module.

[0061] The learning prediction module reads historical normal operation parameters of the geothermal coupled solar heat supply station from the cloud database in combination with corresponding historical meteorological conditions as a data set, divides the data set into a training set and a verification set, and constructs an integrated algorithm prediction model; the integrated algorithm prediction model is trained based on an integrated prediction algorithm of a nearest neighbor node algorithm, a random forest algorithm and reinforcement learning algorithm using the training set, and the integrated algorithm prediction model is optimized using the verification set to predict the user end demand and the photovoltaic system power generation capacity, and the prediction result is transmitted to the real-time decision module;

[0062] The real-time decision module reads the normal operation parameters from the operation checking module and the prediction result from the learning prediction module, respectively, decides the operation condition of each part of the geothermal coupled solar heat supply station by calculating the flow, cold and heat load and power generation capacity, and transmits the decision result to the load adjustment module, and uploads the normal operation parameters from the operation checking module and the prediction result from the learning prediction module to the cloud database;

[0063] The load adjustment module reads the decision result from the real-time decision module, calculates the operation parameters after decision, compares the operation parameters after decision with the fault data in the cloud database, confirms that the power supply system and the heat supply system of the geothermal coupled solar heat supply station can normally operate, and transmits the adjustment signal to the DCS system of the geothermal coupled solar heat supply station; if the power supply system and the heat supply system of the geothermal coupled solar heat supply station cannot normally operate after adjustment, the adjustment signal is ignored to maintain the original working condition, and the error adjustment data is uploaded to the cloud database;

[0064] The cloud database stores normal operation parameters, fault operation parameters, prediction result data and fault decision data and can compare with uploaded data.

[0065] Preferably, the monitoring period of the operation checking module, the decision period of the learning prediction module, the decision period of the real-time decision module and the adjustment period of the load adjustment module are consistent with the preset time step.

[0066] Preferably, the operation checking module monitors the operation parameters of the geothermal coupled solar heating station, including the inlet and outlet water temperature and flow of the user end, the inlet and outlet water temperature and flow of the deep geothermal pipe, the water temperature and flow of the shallow geothermal pipe, the inlet and outlet water temperature and flow of the cooling tower, the temperature and water storage condition of the underground water tank, the power consumption of the geothermal energy system, the power generation of the photovoltaic array and the power storage condition of the energy storage battery, and the operation checking module records the current operation parameters in the form of date, time period, device and working condition; the operation checking module presets an abnormal parameter range, after obtaining the current operation parameters, firstly judges whether the flow, energy and power of the power supply system and the heating system of the geothermal coupled solar heating station are conserved based on the current operation mode, and carries out fault discrimination by comparing the preset abnormal parameter range; when there is possible abnormal data, the operation checking module transmits the current fault data to the cloud database, the cloud database compares the operation parameters of three adjacent time steps and the historical fault database operation parameters, determines the fault position and actively alarms, generates a fault guidance scheme, and uploads the fault data and the fault position to the cloud database;

[0067] The cloud database comprises a normal operation parameter library, a fault operation parameter library, a prediction result database and a fault decision database; the cloud database has a data comparison function, specifically comprising comparing the possible fault data transmitted by the operation checking module with the historical fault data, comparing the decision operation parameters calculated by the load adjustment module with the fault operation database, and comparing the prediction result with the corresponding real data; when the prediction result error or the fault decision in a period of time exceeds a threshold value, the cloud database sends a reminder signal to prompt the maintenance personnel to update and learn the prediction module.

[0068] Compared with the prior art, the present application has the following beneficial technical effects:

[0069] This invention discloses an intelligent control method for geothermal coupled solar heating stations. Based on machine learning principles, it constructs an integrated algorithm prediction model to predict the user's heating and cooling demand and the photovoltaic power generation of the geothermal coupled solar heating station after a certain period based on current meteorological conditions. The method then adjusts the station's operating parameters in advance based on the prediction results. By predicting and matching loads through the integrated algorithm prediction model, the energy utilization rate of the geothermal coupled solar heating station can be improved, and the station's safety can be ensured. The integrated prediction algorithm consists of three parts: the nearest neighbor algorithm, the random forest algorithm, and the reinforcement learning algorithm. The outputs of each algorithm are weighted and averaged to output the final prediction result. Compared with using a single algorithm, by comparing the errors of each part of the three algorithms and updating the weights, the stability and accuracy of the prediction model can be improved. This invention monitors the operating parameters of the geothermal coupled solar heating station in real time through a DCS system. When abnormal operating parameters occur, it actively alarms, identifies the fault location, and provides guidance. This invention also records historical operating parameters of the geothermal coupled solar heating station through a cloud database, providing data for training the prediction model and troubleshooting.

[0070] The intelligent control system for geothermal coupled solar heating stations disclosed in this invention is simple to construct, highly automated, and fully functional. Attached Figure Description

[0071] Figure 1 This is a block diagram illustrating the principle of the intelligent control method for geothermal coupled solar heating stations of the present invention.

[0072] Figure 2 This is a schematic diagram of the intelligent control system for the geothermal coupled solar heating station of the present invention.

[0073] Figure 3 This is a schematic diagram of the integrated algorithm prediction model of the present invention;

[0074] Figure 4 This is a schematic diagram of the geothermal coupled solar heating station in an embodiment of the present invention.

[0075] In the diagram, 1 is the first water pump; 2 is the second water pump; 3 is the third water pump; 4 is the fourth water pump; 5 is the deep underground pipe; 6 is the underground water storage tank; 7 is the shallow underground pipe group; 8 is the closed cooling tower; 9 is the heating device; 10 is the photovoltaic array; 11 is the photovoltaic power generation device; 12 is the energy storage battery; 13 is the first solenoid valve; 14 is the second solenoid valve; 15 is the third solenoid valve; and 16 is the fourth solenoid valve. Detailed Implementation

[0076] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. These descriptions are intended to explain the invention and not to limit it.

[0077] like Figure 4Fig. 1 is a structural diagram of a typical geothermal coupled solar heating station, the system comprising a photovoltaic array 10, a photovoltaic power generation device 11, an underground water storage tank 6, a heating device 9, an energy storage battery 12, a composite ground loop system comprising a shallow ground loop group 7 and a deep ground loop 5 connected in parallel, a closed cooling tower 8; the entire system is also provided with a first water pump 1, a second water pump 2, a third water pump 3, a fourth water pump 4, a first electromagnetic valve 13, a second electromagnetic valve 14, a third electromagnetic valve 15 and a fourth electromagnetic valve 16, and the working conditions during operation are transmitted to the DCS system. The shallow ground loop group 7 is a ground loop with a depth of 100-1000 m, and the deep ground loop 5 is a ground loop arranged at a depth of 1200-3000 m. The shallow ground loops are arranged in a polyline series, and the drilling spacing of the shallow ground loops is 5-10 m; the deep ground loop 5 adopts a casing heat exchanger, which connects the condenser and the evaporator to form a heat exchange unit, and takes heat without taking water.

[0078] As Figure 2 Fig. 2 is a schematic diagram of the intelligent control system of the geothermal coupled solar heating station of the present application, the intelligent control system comprising an operation inspection module, a learning and prediction module, a real-time decision module, a load adjustment module and a cloud database.

[0079] As Figure 1 The main functions of each module and the working principles between the modules are as follows:

[0080] The operation inspection module is used for monitoring the operation of the heating station, and the current operation parameters are diagnosed for fault at a preset time step as a period, and the normal operation parameters are transmitted to the real-time decision module; when there is possible abnormal data, the module transmits the current fault data to the cloud database, compares the historical data through the cloud database to determine the fault condition, actively alarms and gives a fault guidance scheme to help the operation and maintenance personnel to quickly eliminate the fault and ensure the normal operation of the unit.

[0081] The learning and prediction module reads the historical normal operation parameters of the heating station from the cloud and combines the imported corresponding historical meteorological conditions as a data set, and the data set is divided into a training set and a verification set. The module trains a prediction model based on an integrated prediction algorithm combining the nearest neighbor node, the random forest and the reinforcement learning, and optimizes the prediction model using the verification set. The module and the operation inspection module act at the same frequency, and predict at a preset time step as a period, and transmit the prediction results to the real-time decision module.

[0082] Real-time decision module, which reads normal operation parameters from the operation check module and prediction results from the learning prediction module as decision basis, makes decisions on the operation of each part of the heating station by calculating the flow, cold and heat load and power generation, and transmits the decision results to the load adjustment module, uploads the operation parameters of the operation check module and the prediction results of the learning prediction module to the cloud database, and provides decisions at a preset time step as a cycle.

[0083] Load adjustment module, which reads the decision results from the real-time decision module, compares the adjusted operation parameters with the cloud database fault data, confirms that the system can operate normally, and transmits the adjustment signal to the DCS system; if it cannot operate normally after adjustment, it ignores this adjustment signal and maintains the original working condition, and uploads the error decision to the cloud. The load adjustment module adjusts at a preset time step as a cycle.

[0084] Cloud database, including normal operation parameter library, fault operation parameter library, prediction result database, fault decision database, for storing normal operation parameters of heating station, adjustment decisions, fault operation parameters and corresponding decisions. The cloud database transmits and reads corresponding data according to the needs of other modules, and the data comparison function in the cloud database can compare the data provided by other modules with the data in the library and find similar data. When the prediction result error and fault decision in a period of time exceed the threshold, the cloud database sends a reminder signal to prompt the maintenance personnel to update the learning prediction module.

[0085] The geothermal coupling solar heat supply station is divided into a heat supply mode and a refrigeration mode according to an operation mode. When the temperature is high in summer and the cold load is large, the geothermal energy system enters the refrigeration mode, the first electromagnetic valve 13 is closed, the second electromagnetic valve 14 is closed, the third electromagnetic valve 15 is closed, and the fourth electromagnetic valve 16 is opened. The high-temperature water flowing from the user end enters the shallow geothermal pipe for cooling through the fourth water pump 4, and then flows out from the shallow geothermal pipe group 7 into the closed cooling tower 8 for cooling. The low-temperature water is delivered to the user end by the first water pump 1. When the above cooling circuit cannot meet the refrigeration load, the second electromagnetic valve 14 is opened, and the photovoltaic power generation device 11 drives the heating device 9 to cool the underground water storage tank 6. The low-temperature water is delivered to the user end by the third water pump 3. When the temperature is low in autumn and winter and the heat load is large, the geothermal energy system enters the heat supply mode, the first electromagnetic valve 13 is opened, the second electromagnetic valve 14 is closed, the third electromagnetic valve 15 is opened, and the fourth electromagnetic valve 16 is closed. The low-temperature water flowing from the user end is heated through the second water pump 2 and the deep geothermal pipe 5. The deep geothermal pipe 5 takes heat without water, and the medium-deep geothermal energy is guided out through the heat exchange medium. At the same time, the low-temperature water of the user end flows into the shallow geothermal pipe group 7 through the fourth water pump 4, obtains heat, and then flows out. The high-temperature water is delivered to the user end by the first water pump 1. When the above cooling circuit cannot meet the heat supply load, the second electromagnetic valve 14 is opened, the photovoltaic power generation device 11 drives the heating device 9 to heat the underground water storage tank 6, and the high-temperature water is delivered to the user end by the third water pump 3.

[0086] The geothermal coupling solar heat supply station is powered by a photovoltaic power generation system. On the premise of meeting the power demand of the system, the photovoltaic system can participate in grid-connected power generation, and the power that cannot participate in grid-connected peak regulation is stored in the energy storage battery 12. When the energy storage battery 12 is full of electricity, the power that cannot be connected to the grid is used to drive the cooling and heating device to store energy in the underground water storage tank 6. When the photovoltaic power generation device 11 is insufficient, the energy storage battery 12 provides energy for the geothermal energy system. When the energy storage battery 12 is out of energy and the photovoltaic power generation device 11 is insufficient, the geothermal energy system will enter the external power supply mode.

[0087] The operation checking module records the current operation parameters at a preset time step, including the water temperature and flow rate of the user end inlet and outlet, the water temperature and flow rate of the deep buried pipe 5 inlet and outlet, the water temperature and flow rate of the shallow buried pipe 7, the water temperature and flow rate of the cooling tower inlet and outlet, the temperature and water storage condition of the underground water storage tank 6, the power generation of the photovoltaic array 10, and the power storage condition of the energy storage battery 12. The module records the current operation parameters in the form of date, time period, device, and working condition. The module presets abnormal parameter values according to the cloud database historical operation parameters, has a fault discrimination function, and after obtaining the current operation parameters, the module first judges whether the flow rate, heat, and power are conserved based on the current operation mode, compares the preset values to discriminate faults. When there is possible abnormal data, the module transmits the current operation parameters to the cloud database, the cloud database compares the three adjacent time operation parameters with the historical fault database operation parameters to determine the fault condition and actively alarm, generates a fault guidance scheme, and uploads the fault data and fault location to the cloud database.

[0088] The learning prediction module has two functions of training and prediction. The module reads the historical normal operation parameters of the heating station from the cloud and combines the imported corresponding historical meteorological conditions as a data set, which is divided into a training set and a validation set. The module trains a prediction model based on an integrated prediction algorithm combining the nearest neighbor node, random forest, and reinforcement learning, and optimizes the prediction model using the validation set. After obtaining the prediction model, the learning prediction module predicts the cold and heat load and the photovoltaic system power generation after one time step according to the current meteorological conditions. The module and the operation checking module work at the same frequency to read the operation parameters transmitted by the operation checking module at a preset time step for prediction, and transmit the prediction results to the real-time decision module.

[0089] The real-time decision module gives a decision on the entire system according to the prediction value given by the learning prediction module and the current load given by the operation checking module, and uploads the operation checking module operation parameters and the learning prediction module prediction results to the cloud database. Taking winter heating regulation as an example, the specific steps include the following:

[0090] S1: The real-time decision module calculates the flow rate q of each device of the system according to the current operation condition given by the operation checking module i i = 1, 2, 3,..., n;

[0091] The heat provided by each device e i i = 1, 2, 3,..., n;

[0092] The power consumption w of each device i i = 1, 2, 3,..., n;

[0093] where the power generation is positive, the power consumption is negative, and the external power supply is we .

[0094] S2: The real-time decision module makes decisions according to the prediction values given by the learning prediction module. When the user end heat load prediction value, the photovoltaic system power generation prediction value and the operation parameter difference exceed the set threshold value, the decision program is started. The adjustable parts include the opening of each electromagnetic valve, the distribution method of the photovoltaic system power generation, the inlet and outlet flow of the underground water tank and other parameters. In the adjustment process, the following relationship must be met:

[0095]

[0096] Where Q is the user end flow, and E is the user end heat demand.

[0097] S3: The real-time decision module transmits the devices that need to be adjusted and the decisions to the load adjustment module. No signal is transmitted for parts that do not need to be adjusted.

[0098] Preferably, the load adjustment module adjusts the load according to the decisions provided by the real-time decision module. The module first checks the adjusted working conditions, checks whether the working conditions of each part after adjustment are normally running, compares the adjusted working conditions with the fault operation parameter library and the fault decision database in the cloud database, confirms that the system can normally run, transmits the adjustment signal to the DCS system, and uploads the adjustment data to the cloud; if the system cannot normally run after adjustment, ignore this adjustment signal to maintain the original working condition, and upload the error decision to the cloud.

[0099] Preferably, the cloud database includes a normal operation parameter library, a fault operation parameter library, a prediction result database, and a fault decision database. The cloud database transmits and reads data to the intelligent control system modules, specifically including transmitting historical fault data to the operation check module, storing fault data of the operation check module, transmitting historical normal operation parameters to the learning prediction module, storing normal operation parameters and prediction data transmitted by the real-time decision module, and storing decision data of the load adjustment module. The cloud data has a data comparison function, specifically including comparing the fault data transmitted by the operation check module with the historical fault data, comparing the decision data of the load adjustment module with the fault database and the fault decision database. When the prediction result error and the fault decision exceed the threshold value within a period of time, the cloud database sends a reminder signal to prompt the maintenance personnel to update the learning prediction module.

[0100] Preferably, as shown in Figure 3 , the prediction model training in the learning prediction module includes the following steps:

[0101] Step S1: Obtain the operation parameters when the system is normally running through the cloud database, and import the corresponding historical climate conditions. Normalize the operation parameters and historical climate conditions.

[0102] Step S2: Analyze the main factors affecting the heat load, cold load and photovoltaic system power generation using correlation analysis method, and eliminate low-impact factors. Select the processed data as the data set.

[0103] Step S3: For the above data set, use cross-validation method to divide the training set and validation set respectively.

[0104] Step S4: According to the above training set and validation set, build an integrated algorithm prediction model. The integrated algorithm prediction model is composed of three parts: nearest neighbor algorithm (KNN), random forest algorithm (Random Forest) and adaptive boosting algorithm (AdaBoost), and the prediction result is obtained by weighting the prediction results of each part.

[0105] Step S5: The prediction model will update the weighting of each part of the algorithm prediction result according to the deviation between the prediction data and the actual running parameters. When the prediction result and the running parameters appear deviation for several times within a certain time, retrain the prediction model.

[0106] Step S6: Repeat the above steps S3-S5 to obtain the user-side heat demand prediction model, user-side cold demand prediction model and photovoltaic system power generation prediction model.

[0107] Further, step S1 specifically includes the following steps:

[0108] S1.1: The running parameters in normal operation include 8 influencing factors such as selected time point, corresponding time point instantaneous temperature, daily maximum temperature, daily minimum temperature, daily sunshine time, daily air humidity, daily average wind speed and date.

[0109] S1.2: Normalize the data:

[0110]

[0111] Where x i,j represents the jth parameter in the ith group of data; x' i,j represents the normalized parameter of x i,j ; min(x ,j ) represents the minimum value of the jth parameter in each group, max(x ,j ) represents the maximum value of the jth parameter in each group, and the user heat load y i is also initialized in the same way.

[0112] Further, step S2 specifically includes the following steps:

[0113] S2.1: According to the preset time step, the user heat load, the user cold load and the power generation of the photovoltaic array at the starting time point of each time step are selected as the mother sequence by using the grey correlation method, and the time point, the corresponding time point instantaneous temperature, the daily maximum temperature, the daily minimum temperature, the daily sunshine time, the daily air humidity, the daily average wind speed and the date are selected as the sub-sequences.

[0114] S2.2: Temperature, sunshine time and heat load are negatively correlated, and with the increase of these influencing factors, the user heat load will gradually become smaller. In the heat load correlation analysis, the reciprocal operator needs to be multiplied:

[0115] X' i D1=[x' i,1 d1,x' i,2 d2,...,x' i,j d j ]

[0116]

[0117] Wherein, X' i represents the i-th sub-sequence; x' i,j represents the j-th parameter in the i-th group of data; D1 is the reciprocal operator.

[0118] S2.3: The grey correlation degrees of each influencing factor on the user heat demand, the user cold demand and the power generation of the photovoltaic array are obtained by the grey correlation method, and the low correlation degree factors except the time point are removed from the influencing factors. The processed data is used as the data set R, which is represented as:

[0119]

[0120] Further, step S3 specifically includes the following steps:

[0121] S3.1: The data set R processed in step S2 is divided by using the cross-validation method. Here, 5-fold cross-validation is selected, and the data set R is divided into 5 parts, of which 4 parts are used as the training set and 1 part is used as the validation set.

[0122] S3.2: Step S3.1 is repeated 5 times, and each time a different training set is selected, and a total of 20 training sets and 5 validation sets are obtained.

[0123] Further, step S4 specifically includes the following steps:

[0124] S4.1: A nearest neighbor algorithm prediction model is established, and the initial nearest neighbor parameter k is set to 6. The distance measurement uses the following formula:

[0125]

[0126] where d is the Manhattan distance between two points in n-dimensional space, x i,j represents the jth parameter in the ith group of data.

[0127] S4.2: Select Gaussian function as the weighting based on the distance, use mean square error and mean absolute error as the regression loss function, denoted as:

[0128]

[0129] where a, b, c are parameters, and x is the distance between a point in space and the true value.

[0130]

[0131]

[0132] where y i is the true value of the ith parameter, is the predicted value of the ith parameter, and n is the number of dimensions.

[0133] S4.3: Based on the training set and validation set obtained in step S3, train the above nearest neighbor algorithm model, minimize the regression loss function as the target, optimize the values of a, b, c and k of the Gaussian function through the gradient descent algorithm, use the KD tree method to speed up the training, until the loss function reaches the requirement, and obtain the nearest neighbor algorithm model.

[0134] S4.4: Establish a random forest regression prediction model, use the data set R obtained in step S2, use bagging method with replacement sampling, generate N training subsets, and use the out-of-bag data that is not extracted as the validation set.

[0135] S4.5: Use CART method to divide the feature tree, randomly select m influence factors (m≤j) from j influence factors as the split feature value of the current node of the feature tree, and select the minimum criterion of mean square error and mean absolute error as the splitting standard of the feature tree, denoted as:

[0136]

[0137]

[0138] where y i is the true value of the ith parameter, is the predicted value of the ith parameter, and n is the number of dimensions.

[0139] S4.6: For the prediction model of user-side heat load, select the hyperparameters of the random forest model, initially select N tree= 400 feature trees, each feature tree has a maximum depth of 50, and each decision tree has a feature value of m try = 3, and the prediction result is the average value output by each tree. The number of feature trees and the feature value of the decision tree are iteratively calculated and optimized to minimize the error until the error is less than a set threshold, obtaining a random forest model.

[0140] S4.7: Establish an adaptive boosting algorithm prediction model, select the training set obtained in step S3, select n groups of training samples from it, and assign an initial weight w 1i = 1 / n, i = 1, 2, 3,..., n;

[0141] Let W(1) = (w 11 ,w 12 ,w 13 ,…,w 1n ) represent the initial weight of the sample, and select D = 50 to represent the number of learners.

[0142] S4.8: Iteration d = 1, 2, 3,..., D, train the dth weak learner H d (x) when, use the weak learner H d (x) to predict the validation set output regression error ε d , calculate the maximum error E d and the relative error ε di of the sample on the training set, calculate the weight α d of the weak learner in the final learner, and update the weight w(d+1) of the sample according to the weight α d , which is represented as:

[0143] E d = max(|y i -H d (x i )|)

[0144]

[0145]

[0146]

[0147] W(d+1) = (w d+1,1 ,w d+1,2 ,w d+1,3 ,…,w d+1,n )

[0148]

[0149]

[0150] wherein y i is the true value of the i-th parameter.

[0151] S4.9: After D rounds of training, the D weak learners H are obtained. d (x), according to the weak learner weights, the individual weak learners are combined to obtain an enhanced learner h(x), and the weak learner number D is optimized using the validation set data, denoted as:

[0152]

[0153] S4.10: The three trained models are combined with weights to obtain an integrated algorithm prediction model, and the initial weights of the three models are 1 / 3.

[0154] Further, step S5 specifically includes the following steps:

[0155] S5.1: The integrated algorithm prediction model is optimized. The validation set data obtained in step S3 is input, and the trained nearest neighbor algorithm prediction model, random forest regression prediction model, and adaptive boosting algorithm prediction model are used to obtain outputs, respectively. The squared error of each model prediction result and the true value is used as the loss function. The final prediction result is obtained according to the weighted combination of each part of the prediction result, and the initial weights are all 1 / 3. The historical data is used to optimize the weight proportion of each part until the loss function J θ is less than a set threshold:

[0156] D = w1d1 + w2d2 + w3d3

[0157]

[0158]

[0159] wherein D is the output result of the integrated prediction model; d i ,w i are the prediction results and weights of the nearest neighbor algorithm, random forest algorithm, and adaptive boosting algorithm, respectively; w' i is the updated weight; d' is the true value; and k is a parameter for controlling the weight update.

[0160] S5.2: The integrated prediction model uploads the prediction result to the cloud database each time. When the prediction result of the integrated prediction model and the true value have an error that exceeds a set value within a certain time step, the cloud database sends a signal to remind the maintenance personnel to update the learning prediction module, and the prediction model is retrained by repeating the above steps S1-S5.1.

[0161] It should be noted that the above-mentioned only part of the embodiments of the present application, according to the system described in the present application, the equivalent changes are included in the scope of the present application. The skilled in the art of the present application can be described in the specific examples of similar ways to replace, as long as not deviating from the structure of the present application or beyond the scope defined in the claims, are within the scope of the present application.

Claims

1. A method for intelligent control of a geothermal coupled solar heating station, characterized in that, The geothermal coupling solar heat supply station comprises a photovoltaic array (10), a photovoltaic power generation device (11), an underground water storage tank (6), a heating device (9), an energy storage battery (12), a composite ground buried pipe system, the photovoltaic array (10), the photovoltaic power generation device (11) and the energy storage battery (12) are sequentially connected, the photovoltaic array (10), the photovoltaic power generation device (11) and the heating device (9) are sequentially connected, the composite ground buried pipe system comprises a shallow ground buried pipe group (7) and a deep ground buried pipe (5) connected in parallel, and a closed cooling tower (8); the closed cooling tower (8) is connected with a heat user and the shallow ground buried pipe group (7), and transmits the working condition during operation to a DCS system, the shallow ground buried pipes are arranged in a polyline series connection mode, the deep ground buried pipe (5) adopts a casing heat exchanger, and a condenser and an evaporator are connected to form a heat exchange unit, and heat is taken without taking water; comprising: S1: monitoring the operation parameters of the geothermal coupling solar heat supply station according to a preset monitoring period, and performing fault diagnosis on the obtained operation parameters; if the operation parameters are abnormal, the operation parameters are compared with historical data, and a solution to the fault is determined according to the comparison result; if the operation parameters are normal, S2 is turned to; S2: combining historical normal operation parameters of the geothermal coupling solar heat supply station with corresponding historical meteorological conditions as a data set, dividing the data set into a training set and a verification set, and constructing an integrated algorithm prediction model; using the training set, an integrated prediction algorithm combining a nearest neighbor node algorithm, a random forest algorithm and an enhanced learning algorithm is used to train the integrated algorithm prediction model, and the verification set is used to optimize the integrated algorithm prediction model, and the user end demand and the photovoltaic system power generation capacity are predicted; in S2, the training method of the prediction model comprises: S2.1: according to the historical normal operation parameters of the geothermal coupling solar heat supply station, the corresponding historical meteorological conditions are imported, and the historical normal operation parameters and the historical meteorological conditions are normalized; S2.2: using a correlation analysis method to analyze the main factors affecting the heat load, the cold load and the photovoltaic system power generation capacity, and removing low-influence data, the remaining data are used as a data set; S2.3: using a cross-validation method to divide the data set into a training set and a verification set; S2.4: constructing an integrated algorithm prediction model according to the training set and the verification set, respectively training a nearest neighbor algorithm prediction model, a random forest algorithm prediction model and an adaptive boosting algorithm prediction model, and outputting the final prediction result by weighting the prediction results; S2.5: the integrated algorithm prediction model updates the corresponding weight according to the prediction results of the nearest neighbor algorithm prediction model, the random forest algorithm prediction model and the adaptive boosting algorithm prediction model and the deviation of the actual operation parameters of the geothermal coupling solar heat supply station; when the prediction results of the integrated algorithm prediction model and the operation parameters deviate continuously for a plurality of times within a certain time, the integrated algorithm prediction model constructed in S2.4 is retrained; S2.6: repeating steps S2.3-S2.5, respectively obtaining a user end heat demand prediction model, a user end cold demand prediction model and a photovoltaic system power generation capacity prediction model; S3: According to the normal operation parameters obtained in S1 and the prediction results of S2, the operation conditions of each part of the geothermal coupled solar heat supply station are decided by calculating the checked flow, cold and heat load and power generation; S4: According to the decision results obtained in S3, the operation parameters after decision are calculated, the operation parameters after decision are compared with the fault data, and after confirming that the power supply system and heat supply system of the geothermal coupled solar heat supply station can operate normally, the adjustment signal is transmitted to the DCS system of the geothermal coupled solar heat supply station; If the power supply system and heat supply system of the geothermal coupled solar heat supply station cannot operate normally after adjustment, ignore this adjustment signal and maintain the original working condition, and upload the error adjustment data.

2. The intelligent control method of geothermal coupled solar heat supply station according to claim 1, characterized in that, In S1, the operation parameters include selected time point, corresponding time point instantaneous temperature, daily maximum temperature, daily minimum temperature, daily sunshine duration, daily air humidity, daily average wind speed and date; The operation parameters are normalized before fault diagnosis: wherein denotes the jth parameter in the ith group of data; denotes the normalized parameter; denotes the minimum value of the jth parameter in each group, denotes the maximum value of the jth parameter in each group.

3. The intelligent control method of geothermal coupled solar heating station according to claim 1, wherein, S2.2 specifically includes: S2.2.1: According to the preset time step, the user heat load, user cold load and power generation of photovoltaic array at the start time point of each time step are selected as the mother sequence by using the grey correlation method; The subsequence selection time point, corresponding time point instantaneous temperature, daily maximum temperature, daily minimum temperature, daily sunshine duration, daily air humidity, daily average wind speed and date; S2.2.2: According to the negative correlation between temperature, sunshine duration and heat load, when analyzing the correlation of heat load, multiply the reciprocal operator: wherein, denotes the i-th sub-sequence; denotes the j-th parameter in the i-th group of data; is the reciprocal operator; S2.2.3: The grey correlation degrees of each influencing factor on user end heat demand, user end cold demand and photovoltaic array power generation are obtained by using the grey correlation method, and the low correlation degree factors except time point are removed from the influencing factors; And the processed data is used as data set R, which is represented as: 。 4. The intelligent control method of geothermal coupled solar heat supply station according to claim 1, characterized in that, S2.3 specifically includes: S2.3.1: The data set R obtained by processing in step S2.2 is divided by using cross validation method, and 5-fold cross validation is selected, that is, the data set R is divided into 5 parts, of which 4 parts are used as training set and 1 part is used as validation set; S2.3.2: Repeat step S2.3.1 for 5 times, select different training set each time, and get 20 training sets and 5 validation sets.

5. The intelligent control method of geothermal coupled solar heating station of claim 1, wherein, S2.4 specifically includes: S2.4.1: Establish the nearest neighbor algorithm prediction model, set the hyperparameter k, select Manhattan distance as the distance measure, select mean square error and mean absolute error as the loss function, select Gaussian function as the weighting based on distance proximity, based on the training set and validation set obtained in step S2.3, minimize the loss function as the target, use KD tree to accelerate the training, and get the nearest neighbor algorithm prediction model; It is represented as: wherein, is the true value of the i-th parameter, is the predicted value of the i-th parameter, n is the number of dimensions; Wherein, a, b, c are parameters, x is the distance between a certain point in space and the true value; S2.4.2: Establish a random forest regression prediction model, initially select a feature tree, set the maximum depth of each feature tree to 50, and initially select the feature value of each feature tree to be ; select the data set R obtained in step S2.2, use the bagging method to sample with replacement to generate training subsets, and use the unselected out-of-bag data as the validation set; use the CART method to divide the feature tree, randomly select impact factors (j ≤j) as the split feature value of the current node of the feature tree, select the minimum mean square error and average absolute error minimization criterion as the split criterion of the feature tree, and the prediction result is the mean value output by each tree; iteratively calculate the number of feature trees and the feature values of the decision trees to optimize the error minimization target until the error is less than the set threshold, and obtain the random forest prediction model; S2.4.3: Establish an adaptive boosting algorithm prediction model. Select the training set obtained in step S2.3, choose n training samples from it, and assign initial weights to the training samples. ,use The initial weights of the samples are represented by , and D represents the number of learners; iterations are performed. Train the d-th weak learner When using weak learners Prediction validation set output regression error Calculate the maximum error of the samples on the training set. and relative error Calculate the weights of the weak learner in the final learner. According to weight Update sample weights After training for D rounds, we obtain D groups of weak learners. The reinforcement learner is obtained by combining the weights of the weak learners. And optimize the number of weak learners D using validation set data; denoted as: wherein, is the true value of the i-th parameter.

6. The intelligent control method of geothermal coupled solar heat supply station of claim 1, wherein, S2.5 specifically includes: S2.5.1: optimizing the ensemble algorithm prediction model; inputting the validation set data obtained in step S2.3, using the trained nearest neighbor algorithm prediction model, random forest regression prediction model and adaptive boosting algorithm prediction model to obtain outputs respectively, taking the mean square error of the ensemble algorithm model prediction result and the true value as the loss function, using historical data to optimize the weighting proportion of each part until the loss function is less than a set threshold value: wherein, is the prediction model output result of the ensemble algorithm; , are the prediction results and weights of the nearest neighbor algorithm, the random forest algorithm, and the adaptive boosting algorithm, respectively; is the updated weight; is the true value; is the parameter for controlling the weighted update; S2.5.2: The prediction result of each integrated algorithm prediction model is uploaded to the cloud database, and when the prediction result error or fault decision in a period of time exceeds the threshold value, the cloud database sends a signal to remind the maintenance personnel to update the learning prediction module, and the prediction model is retrained by repeating the above steps S2.1-S2.5.

1.

7. A geothermal coupled solar heat supply station intelligent control system, characterized in that, It includes: The operation checking module monitors the operation parameters of the geothermal coupled solar heat supply station according to a preset monitoring period, and performs fault diagnosis on the obtained operation parameters; If the operation parameter is abnormal, the operation parameter is uploaded to the cloud database; If the operation parameter is normal, the operation parameter is transmitted to the real-time decision module; the operation checking module monitors the operation parameters of the geothermal coupled solar heat supply station, including the inlet and outlet water temperature and flow of the user end, the inlet and outlet water temperature and flow of the deep geothermal pipe, the water temperature and flow of the shallow geothermal pipe, the inlet and outlet water temperature and flow of the cooling tower, the temperature and water storage of the underground water tank, the power consumption of the geothermal energy system, the power generation of the photovoltaic array and the power storage of the energy storage battery, and records the current operation parameters in the form of date, time period, device and working condition; the operation checking module presets an abnormal parameter range, and after obtaining the current operation parameters, first judges whether the flow, energy and power of the power supply system and the heat supply system of the geothermal coupled solar heat supply station are conserved based on the current operation mode, and performs fault discrimination by comparing the preset abnormal parameter range; when there is possible abnormal data, the operation checking module transmits the current fault data to the cloud database, the cloud database compares the operation parameters of three adjacent time steps with the historical fault database operation parameters, determines the fault position and actively alarms, generates a fault guidance scheme, and uploads the fault data and fault position to the cloud database; The cloud database includes a normal operation parameter library, a fault operation parameter library, a prediction result database and a fault decision database; The cloud database has a data comparison function, which specifically includes comparing the possible fault data transmitted by the operation checking module with the historical fault data, comparing the decision operation parameters calculated by the load adjustment module with the fault operation database, and comparing the prediction results with the corresponding real data; when the prediction result error or fault decision in a period of time exceeds the threshold value, the cloud database sends a reminder signal to prompt the maintenance personnel to update and learn the prediction module; The learning prediction module reads the historical normal operation parameters of the geothermal coupled solar heat supply station from the cloud database in combination with the corresponding historical meteorological conditions as a data set, divides the data set into a training set and a validation set, and constructs an integrated algorithm prediction model; the training set is used to train the integrated algorithm prediction model based on the integrated prediction algorithm of the nearest neighbor node algorithm, the random forest algorithm and the reinforcement learning algorithm, and the validation set is used to optimize the integrated algorithm prediction model, the user end demand and the photovoltaic system power generation capacity are predicted, and the prediction result is transmitted to the real-time decision module; The real-time decision module reads the normal operation parameters from the operation checking module and the prediction results from the learning prediction module, respectively, decides the operation conditions of each part of the geothermal coupled solar heat supply station by calculating and checking the flow, cold and heat load and power generation capacity, and transmits the decision result to the load adjustment module, and uploads the normal operation parameters from the operation checking module and the prediction results from the learning prediction module to the cloud database. The load adjustment module reads the decision result from the real-time decision module, calculates the operation parameter after the decision, compares the operation parameter after the decision with the fault data in the cloud database, and transmits the adjustment signal to the DCS system of the geothermal coupling solar heat supply station after confirming that the power supply system and the heat supply system of the geothermal coupling solar heat supply station can normally operate. If the power supply system and the heat supply system of the geothermal coupling solar heat supply station cannot normally operate after adjustment, the adjustment signal is ignored to maintain the original working condition, and the error adjustment data is uploaded to the cloud database. The cloud database stores normal operation parameters, fault operation parameters, prediction result data and fault decision data and can compare the uploaded data.

8. The geothermal coupled solar heating station intelligent control system of claim 7, wherein, The monitoring period of the operation checking module, the decision period of the learning prediction module, the decision period of the real-time decision module and the adjustment period of the load adjustment module are consistent with the preset time step.

Images