Research on CCHP System Grid Connection Interface Device Based on Elman Neural Network

With the promotion of the concepts of “energy Internet” and “multi energy complementation”, the combined cold heat and power (CCHP) system technology has been paid more and more attention by the energy power circles at home and abroad. CCHP system generates electricity through internal combustion engine, meanwhile, the waste heat discharged after power generation is supplied with heat and cooling to users through waste heat recovery equipment, so as to effectively improve the energy utilization rate. However, with the development of information and intelligence in the power grid, the threat of various network malicious attacks on the power industrial control terminal is growing. The traditional CCHP system grid connection interface device operating environment is open and vulnerable to network malicious attacks. In this paper, an interface device of CCHP based on Elman neural network algorithm is proposed. The technical problem to be solved is to be able to realize active defense against unknown attacks and improve the security and operation reliability of CCHP system.


Introduction
The CCHP system [1-4] is a cogeneration system based on the concept of energy cascade utilization, using natural gas as a primary energy source to generate three types of energy: cold, heat and electricity. It uses natural gas as fuel. The high-temperature flue gas obtained by burning small natural gas using small gas turbines, gas internal combustion engines, and micro-turbines is first used for power generation, and then uses waste heat for heating in winter; it is cooled by a refrigerator in summer; can provide domestic hot water, making full use of exhaust heat. CCHP system's energy utilization rate can be as high as 80%, which saves a lot of energy. Compared with the traditional centralized energy supply method, the combined cooling, heating, and power system not only has the advantages of high energy efficiency, clean and environmental protection, good safety, peak-cutting and valley-filling, and good economic benefits. It can also meet the end-user's diverse energy supply. Demand is an important development direction of the power industry and energy industry.
In recent years, CCHP system technology has received increasing attention from domestic and foreign energy and power industries. It has developed rapidly in domestic and foreign markets, and engineering application technology has become increasingly mature. On the one hand, because the end-user's energy demand is affected by comprehensive factors such as seasonal changes, building types, and power supply of the power grid, the CCHP system needs to accurately and timely connect the relevant status data of cold, heat, and electricity to the grid when it is connected to the grid. The interface device is uploaded to the higher-level dispatch center; but on the other hand, as the power grid gradually moves towards informationization and intelligence, power industrial control terminals are increasingly threatened by various network malicious attacks, and the current operating

Elman Neural Network Training Model
The Elman neural network [9][10][11][12] model training process, as shown in figure 1, specifically includes: (1) The sigmoid activation function is used to normalize each hidden element to become the probability value that they are in activation: The sigmoid excitation function is: Where e is a constant with a value of 2.718, and x indicates that the value transmitted by the explicit layer is multiplied by the connection weight plus the deviation of the hidden layer; the value transmitted by the explicit layer is the sample data in the input sample, and the connection weight is neural. Links between the yuan; (2) Calculate the probability that the hidden element is activated: (2) Where P ℎ = 1 v refers to the probability of the hidden element being activated, ℎ is the jth activated hidden element, 0 represents the initial display layer, represents the i-th display element, and each of the i displays is The value of the sample data of the training samples in the training database, w n is the weight of the connection between the explicit layer (input layer, inherited layer) and the hidden layer, m is the total number of explicit elements in the explicit layer, and n is The total number of hidden elements in the hidden layer, j is the j-th hidden element, c j is the offset of the j-th hidden element, and the initial offset c of the hidden layer is 0; between the explicit layer and the hidden layer The connection weight is the connection between the explicit layer and the hidden layer, and the initialization of the connection weight w comes from a random number with a normal distribution N (0, 0.01); (3) Calculate the probability that the explicit element is activated: (3) Where P = 1 ℎ 0 is the probability that the display element is activated, v i represents the i-th activated element after reconstruction, h 0 represents the hidden layer, h j represents the j-th hidden element, h j W n×m is the connection weight between the explicit layer and the hidden layer, n is the total number of hidden elements in the hidden layer, i is the i-th explicit element, and b is the offset of the explicit element, which is initialized to b i =log the training sample whose i-th feature is activated, the feature is sample data in the sample, and the hidden layer represents the activated feature.
(4) Then calculate the probability that the hidden element is activated again with the reconstructed explicit element, and obtain a new hidden layer h ', that is, use the value of the hidden element after activation in step (2) as the input value and the corresponding Multiply the connection weights, and then sum these products before adding the deviations. The result is the reconstructed value, which is the approximate value of the original input. Then use the formula in step (3) to calculate the probability of activation of the display element. Then, Map the reconstructed activation explicit cell value to the hidden cell; Where P h j ′ = 1 v ′ refers to the probability of the hidden element being activated, h j ′ represents the j-th hidden element activated by the reconstruction layer after reconstruction, and v ′ represents the reconstruction Layer, v i ′ represents the i-th display element, the i-th display is the value of the reconstructed display element, W n×m is the connection weight between the display layer and the hidden layer, and m is the display The total number of explicit elements in the layer, n is the total number of hidden elements in the hidden layer, j is the j-th hidden element, and c j is the offset of the jth hidden element; the connection between the explicit layer and the hidden layer The weight is the connection between the explicit layer and the hidden layer.
(5) Update the offset weights. The update formula is: Where △W represents the difference (Error feedback) between the display element and the input value after reconstruction, α is the learning efficiency value of 0.01, and W is the connection weight before the update (W's initialization comes from the normal distribution N (0 , 0.01) random number), 0 is the initial explicit layer assigned to the explicit element, 0 is the transpose of 0 , ℎ 0 is the hidden layer activated after the explicit layer mapping, v ′ is the reconstruction The reconstructed explicit layer maps the value to the hidden layer, and the hidden layer is activated to obtain h ′ ; the initial visible layer is the sample data in the sample; (6) After sufficient training of an initial manifestation is completed, determine the connection weight and offset of the training sample data; (7) Repeat (1)-(6) until all the sample data are trained, and finally get the trained Elman neural network model.

Introduction Of CHP Grid-connected System
As shown in Figure 2, the CCHP grid-connected system structure includes a CCHP control management unit, a CCHP grid-connected interface device [13][14][15][16], and a superior dispatch center. In the figure, the solid line is the energy line, the dotted line is the communication line, and the gridconnected interface device is a communication bridge that connects the upper-level dispatch center and the lower-level control management unit. The control management unit obtains electricity from the generator, refrigeration unit, heat exchanger, and domestic hot water. And cold and hot status data and according to the relevant remote adjustment, start and stop commands sent by the superior dispatch center via the grid-connected interface device, while the grid-connected interface device receives status data (such as gas turbine power, power generation capacity) sent by the control management unit , power quality, heat supply, cooling capacity information and grid connection point voltage, current, power and other data, among which gas turbine power, grid connection point voltage, current, power and other data are susceptible to cyber attacks) (control system, temperature monitoring device and grid connection point) collects voltage, current, active power and reactive power, gas turbine power, cooling capacity, heat and other information, and receives relevant remote signal data from the lower-level controller. Generator or higher-level dispatch center) in real time to convey information such as gas turbine power, power generation, power quality, heat supply, and cooling capacity, as well as Point voltage, current, power and other data.  Figure 2. CCHP grid connection system structure

Structure and Strategy of Grid Connection Interface Devicel
The CCHP system grid connection interface device based on Elman neural network includes measurement module, control module, attack detection module, power module, communication module, display module, output module, input module and memory. As shown in Figure 3.  Figure 3. Structure diagram of interface device The attack detection module is respectively connected with the input module, the measurement module, the power module and the communication module, and the AD conversion module is used for AD conversion between the measurement module and the attack detection module; The control module is respectively connected with the power module, the display module, the output module, the memory and the communication module; The measurement module is used to measure some state data in CCHP system data stream, and then send it to attack detection module after AD conversion; The control module is used to communicate with the upper layer and the lower layer through the communication module, receive the alarm sent by the attack detection module, and send the alarm to ICNISC 2020 Journal of Physics: Conference Series 1646 (2020) 012094 IOP Publishing doi:10.1088/1742-6596/1646/1/012094 6 the display module for display. The control module also sends the output signal to the output module; the output signal includes the output control common connection point switch, CCHP system intercooler, thermal load equipment switch, breaker switch command signal The attack detection module is respectively connected with the input module, the measurement module, the power module and the communication module, and the AD conversion module is used for AD conversion between the measurement module and the attack detection module; The control module is respectively connected with the power module, the display module, the output module, the memory and the communication module; The measurement module is used to measure some state data in CCHP system data stream, and then send it to attack detection module after AD conversion; The control module is used to communicate with the upper layer and the lower layer through the communication module, receive the alarm sent by the attack detection module, and send the alarm to the display module for display. The control module also sends the output signal to the output module; the output signal includes the output control common connection point switch, CCHP system intercooler, thermal load equipment switch, breaker switch command signal ; Further, the control module is also used to send alarms and log records to the upper layer and / or to the memory for storage through the communication module.
Further, before the attack detection module inputs the characteristic data flow into the Elman neural network model for real-time detection and classification, it also needs to train the Elman neural network model, and the training is realized in the following ways: (1) Input samples to Elman neural network model; the samples include positive samples and negative samples, positive samples are normal data that can indicate whether CCHP system grid connection interface device is attacked; negative samples are obtained after network attack on normal data in positive samples; (2) The Elman neural network model was trained repeatedly. Through the detection and analysis of the cold, hot and electricity related state data through Elman neural network, the abnormal data in the state data which is attacked by the network can be found through the active identification [16][17][18][19][20], and the corresponding alarm prompt can be sent after the classification according to the characteristics of the network attack, and the active immunity can be realized by intercepting the abnormal data to avoid the abnormal data.
The attacker further invades the superior dispatching center through the security vulnerability of the grid connection interface device, so as to improve the security and operation reliability of the CCHP system.
Schematic diagram of neural network model is shown in Figure 4.

Conclusions
With the development of information and intelligence in the power grid, the power industry control terminal is threatened more and more by various network malicious attacks. At present, the operation environment of the grid connection interface device is open and vulnerable to various network attacks. These attacks will endanger the confidentiality, integrity and availability of information. Therefore, for the grid connection interface devices that need to be incorporated into the power grid, it is urgent to improve their active immune ability, so as to avoid network attacks to interfere with the normal operation of the power grid and damage the whole CCHP system by invading the grid connection interface devices.
In view of the current network attack risk, this paper proposes an attack identification method based on Elman neural network and grid connection interface device, which can realize the active defense of unknown attacks, improve the security and operation reliability of CCHP system, and then improve the reliability and security of grid connection.

Acknowledgments
This work was by Supported by the National Key R&D Program of China (2018YFB0904900, 2018YFB0904903).