Substation fire early warning method and system and storage medium

Abstract

The application discloses a substation fire identification method based on heat release ion detection, first, an identification model of heat release ion is constructed, particle data received by sensors in the substation are collected, and feature extraction is performed on the data, heat release ion particles are screened out by continuously adjusting internal parameters and iteration times of the model, and influence of environmental gas fluctuation on fire identification is excluded; next, historical data of the screened heat release ion concentration are input into a prediction model, and a change curve of the heat release ion concentration in the substation in the next 1-3 hours is obtained; finally, the predicted heat release ion concentration is compared with a heat release ion concentration curve diagram when a fire occurs, whether there is a fire hazard in the substation is judged, and if there is a fire hazard, a warning signal is sent. The application can effectively improve fire detection accuracy, reduce false positive rate and reduce reaction time, improve warning speed, and reflect fire change in the substation in real time.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of fire monitoring, and particularly relates to a substation fire warning method and system and a storage medium. BACKGROUND

[0002] In recent years, fire accidents caused by electrical faults of equipment in outdoor places of substations occur frequently, but it is difficult to discover and take corresponding measures to deal with the fire at the initial stage due to imperfect fire monitoring and warning measures. Once the electrical equipment catches fire, the fire spreads quickly and is difficult to extinguish, which not only directly burns a large number of primary and secondary equipment, but also causes the nearby equipment to catch fire and explode, resulting in a long time of power outage for repair of the substation and seriously affecting the operation safety of the power grid. At present, there is no very effective method for monitoring the fire in the outdoor places of substations in China. The fire monitoring technology in large spaces is a weak point in the outdoor fire monitoring of substations. Since the flame burning matter spreads in the air is affected by the clearance height and the site area, only when the fire develops to a certain extent, the conventional smoke and temperature sensors will respond, which is difficult to effectively play a role. Moreover, the measurement values of the smoke and temperature sensors are disturbed by the environment and climate, and the real-time measurement data cannot reflect the future trend of the fire. SUMMARY

[0003] The purpose of the present application is to solve the above-mentioned problems existing in the prior art, and to provide a substation fire warning method, system and storage medium.

[0004] The technical solution of the present application is: a substation fire warning method, which is characterized in that:

[0005] A substation fire warning method, which is characterized in that:

[0006] Step one: constructing a recognition model of thermal release ions according to a long short-term memory neural network;

[0007] Step two: collecting particle data in the substation through sensors placed at each point in the substation;

[0008] Step three: extracting features from the collected data;

[0009] Step four: continuously adjusting the internal parameters of the model and the number of iterations;

[0010] Step five: evaluating whether the gas fluctuation belongs to thermal release ions according to the difference between the provided data and the thermal release ion particles;

[0011] Step six: inputting the screened thermal release ion historical concentration data into the established thermal release ion concentration prediction model to predict the concentration data of the thermal release ions in the next 1-3 hours;

[0012] Step seven: compare the predicted data with the concentration fluctuation curve of the thermal dissociation ion when the fire occurs, if the predicted curve is consistent with the thermal dissociation ion curve when the fire occurs and also presents an exponential form fluctuation, it is determined that there is a fire hazard in the future, and the system sends a warning signal to the patrol personnel of the substation to remind the patrol personnel that a fire will occur.

[0013] Further, in step 1, the identification model of the thermal dissociation ion includes a data acquisition layer, an input layer, a hidden layer, a network training layer, a dropout layer, a full connection layer, a softmax layer and an output layer; wherein,

[0014] The data acquisition layer is used to collect and process the particle data received by the sensor in the substation;

[0015] The input layer is used to convert the processed data into the format of network input;

[0016] The network training layer is used to train the model;

[0017] The dropout layer is used to prevent network overfitting;

[0018] The full connection layer is used to extract features from the input data;

[0019] The hidden layer is used to place the feature data obtained after the full connection layer performs feature extraction;

[0020] The softmax layer is used to classify the input data according to the feature extraction results and select the data with the smallest difference from the feature information;

[0021] The output layer is used to output the final data obtained after classification and screening.

[0022] Further, in step two, the particle data in the substation is collected, which includes smoke particle data, thermal dissociation ion data and dust particle data. A group of particle data is collected every 3-5 minutes for storage, and the collection time is 2-3 hours. If there are missing data in multiple collection point data, they are filled in or repeated data are deleted.

[0023] Further, the specific process of feature extraction of the collected data in step three is:

[0024] For the feature information of the collected data, the particle diameter size S, the light transmittance P and the particle density f are selected for analysis:

[0025]

[0026] The output result y is the collected particle data, and it is consistent with the feature information data of the thermal dissociation ion The difference of the comparison is e:

[0027]

[0028] Meanwhile, in order to improve the training effect and enhance the approximation ability of the algorithm model, an additional hidden layer is added after the input layer: the input layer has 3 nodes corresponding to three characteristic information, i.e. particle diameter S, light transmittance P and particle density f; the first, second and third hidden layers have 30-60 nodes respectively, corresponding to an equal number of groups of sampling point data; the output layer has one node corresponding to the calculated difference e;

[0029] The activation function of the first layer of hidden layer nodes is a linear function, and its expression is

[0030]

[0031] The activation function of the second and third layers of hidden layer nodes is a Sigmoid function, and its expression is

[0032]

[0033] wherein each node corresponds to a neuron, the connection weight between the neurons of the input layer and the neurons in the first hidden layer is , the connection weight between the neurons in the first hidden layer and the neurons in the second hidden layer is the same as the connection weight between the neurons in the second hidden layer and the neurons in the third hidden layer, i.e. , and the connection weight between the neurons in the third hidden layer and the neurons in the output layer is , then the input of the neurons in the first hidden layer is:

[0034] The output of the neurons in the first hidden layer is:

[0035]

[0036] The input of the neurons in the second hidden layer is:

[0037]

[0038] The output of the neurons in the second hidden layer is:

[0039]

[0040] The input of the neurons in the third hidden layer is:

[0041]

[0042] The output of the neurons in the third hidden layer is:

[0043]

[0044] The output of the output layer neuron is

[0045] .

[0046] Further, the internal parameters of the model and the number of iterations are continuously adjusted in step four, and the specific process is: adjusting the learning rate and the number of iterations according to the set value, obtaining the weight W, the loop weight A and the bias b, and respectively adjusting the hyperparameters of the corresponding input gate t, the forgetting gate f, the candidate gate g and the output gate j using the gate activation function, wherein the internal learning weight of the network is:

[0047]

[0048]

[0049]

[0050] The process of iteration is:

[0051]

[0052]

[0053] wherein, is a gate activation function, using a sigmoid function, that is, , is a state activation function, using a tanh function;

[0054] The forgetting gate outputs a vector between 0 and 1 by using the information of and , and the 0-1 value in the vector indicates which information in the input network model is retained and which information needs to be discarded; 0 indicates not to retain, and 1 indicates to retain all;

[0055]

[0056] The input gate is used to update the unit state, and the information of the previous hidden state and the current input information is first input to the sigmoid function;

[0057]

[0058] The data is input into the model for model training, and two training termination indicators and are set, and if the training termination indicators and If any one of the values of the two training termination indicators is less than a set value, the internal hyperparameters of the model are readjusted for model training until both values of the two training termination indicators meet the requirements, wherein the calculation formula of the model training termination indicator is:

[0059] wherein is the Euclidean distance, is the number of screened data, is the maximum number of iterations, is the current number of iterations.

[0060] Further, the evaluation process of the thermolysis ion in step five is:

[0061] If the difference e between the collected particle data y and the characteristic information data of the thermolysis ion is within the allowable error range, the particle is determined to be a thermolysis ion, and the historical data of the particle is retained; the allowable error range is If the difference e between the collected particle data y and the characteristic information data of the thermolysis ion exceeds the allowable error range, the particle is determined to be not a thermolysis ion, and the historical data of the particle is removed.

[0062] Further, the specific process of step six is:

[0063] The fitting equation is:

[0064]

[0065]

[0066] Let be the sampling time, wherein i is 1, 2,..., 120,

[0067] be the corresponding historical thermolysis ion concentration, be the fitting value of the thermolysis ion concentration, i.e., the predicted future 1-3 hour thermolysis ion concentration data;

[0068] Calculate the variance size:

[0069] If Q≤1.05, go to step seven, otherwise return to step two.

[0070] ​​​Further, the predicted curve in step seven matches the heat release ion curve when the fire occurs, which means that the difference between the concentration of heat release ions and the concentration value data of heat release ions when the fire occurs is not more than 7% of the concentration value of heat release ions when the fire occurs.

[0071] A substation fire warning system based on heat release ion detection, characterized in that it comprises:

[0072] A heat release ion recognition model construction module is configured to construct a heat release ion recognition model comprising a data collection layer, an input layer, a hidden layer, a network training layer, a dropout layer, a full connection layer, a softmax layer, and an output layer according to a long short-term memory neural network.

[0073] A particle data collection module is configured to collect particle data in the substation according to sensors placed at each point in the substation.

[0074] A feature extraction module is configured to perform feature extraction according to the data collected by the particle data collection module.

[0075] A parameter and iteration number adjustment module is configured to input data into the model for model training until the training termination index value meets the requirements.

[0076] A heat release ion evaluation module is configured to evaluate whether the gas fluctuation belongs to heat release ions according to the difference between the provided data and the heat release ion particles.

[0077] A heat release ion concentration data prediction module is configured to input the screened heat release ion historical concentration data into the established heat release ion concentration prediction model to predict the concentration data of heat release ions in the next 1-3 hours.

[0078] A fire hazard determination and warning module is configured to determine that there will be a fire in the future if the predicted curve matches the heat release ion curve when the fire occurs and also shows exponential form fluctuation, and to send a warning signal to the patrol personnel of the substation.

[0079] A readable computer storage medium storing a computer program, characterized in that the program is executed by a processor to implement the method described in any one of the above.

[0080] The beneficial effects of the present application are: the present application establishes a thermal release ion concentration prediction model, and the method for predicting the change curve of the thermal release ion concentration input into the model in the future few hours by continuously adjusting the internal parameters of the model and the iteration times to monitor the real-time situation of the substation so as to give an early warning of the fire in the first time. The present application improves the fire monitoring and early warning, and proposes a method for rapidly identifying the thermal release ions existing in the air after the fire. The technology does not react to dust, haze, smoke particles and sand, and does not give an early warning, strictly distinguishes the thermal release ions from these, can effectively and accurately detect the concentration of the thermal release ions, only identifies and analyzes the thermal release ions and then decides whether to give an early warning, so that the false alarm is effectively prevented, each detection is reliable, the fire detection accuracy is effectively improved, the false alarm rate is reduced, the reaction time is reduced, the early warning speed is improved, and the change of the fire in the substation is reflected in real time. BRIEF DESCRIPTION OF DRAWINGS

[0081] Figure 1 is the algorithm flowchart of the present application;

[0082] Figure 2 is the corresponding relationship diagram between neurons of each layer;

[0083] Figure 3 is the concentration fluctuation curve diagram of the thermal release ions when the fire occurs;

[0084] Figure 4 is the comparison diagram of the predicted value and the actual value of the thermal release ion concentration when the fire occurs;

[0085] Figure 5 is the system structure diagram of the present application. DETAILED DESCRIPTION

[0086] As shown in Figure 1 , the present application discloses a substation fire identification method based on thermal release ion detection, and the steps are as follows:

[0087] A substation fire early warning method, and the steps are as follows:

[0088] Step 1: build a thermal release ion identification model;

[0089] According to the long short-term memory neural network, the thermal release ion identification model is built, and the thermal release ion identification model includes a data acquisition layer, an input layer, a hidden layer, a network training layer, a dropout layer, a full connection layer, a softmax layer and an output layer; wherein,

[0090] The data acquisition layer is used for acquiring the particle data received by the sensor in the substation and processing the particle data;

[0091] The input layer is used for converting the processed data into the format of network input.

[0092] a network training layer for training the model;

[0093] a dropout layer for preventing overfitting of the network;

[0094] a fully connected layer for feature extraction of the input data;

[0095] a hidden layer for placing the feature data obtained after the feature extraction of the fully connected layer;

[0096] a softmax layer for classifying the input data according to the result of the feature extraction, and selecting the data with the smallest difference from the feature information;

[0097] an output layer for outputting the data finally obtained after the classification and screening;

[0098] Step two: collecting the particle data received by the sensor in the substation;

[0099] The particle data in the substation is collected by the sensors placed at each point in the substation, and the particle data includes smoke particle data, thermal ion data and dust particle data. One set of particle data is collected and stored every three minutes, and the collection time is 2 hours, a total of forty sets of sampling point data are collected and stored. If there are missing data in the forty sets of collected data, the missing data is filled or the repeated data is deleted.

[0100] Step three: feature extraction of the collected data;

[0101] According to the feature information of the collected data, the particle diameter size S, the light transmittance P and the particle density f are selected for analysis:

[0102] According to the feature information of the collected data, the particle diameter size S, the light transmittance P and the particle density f are selected for analysis:

[0103]

[0104] The output result y is the collected particle data, and it is compared with the thermal ion feature information data The difference of comparison is e:

[0105]

[0106] As Figure 2The hidden layer is added after the input layer to improve the training effect and enhance the approximation ability of the algorithm model. The input layer has three nodes corresponding to three characteristic information, i.e., particle diameter S, light transmittance P and particle density f. The first, second and third hidden layers have 40 nodes corresponding to 40 groups of sampling point data. The output layer has one node corresponding to the calculated difference e.

[0107] The activation function of the first hidden layer node is a linear function, and the expression is

[0108]

[0109] The activation function of the second and third hidden layer nodes is a Sigmoid function, and the expression is

[0110]

[0111] wherein each node corresponds to a neuron, the connection weight between the neuron of the input layer and the neuron in the first hidden layer is the connection weight between the neuron in the first hidden layer and the neuron in the second hidden layer is the same as the connection weight between the neuron in the second hidden layer and the neuron in the third hidden layer, i.e. the connection weight between the neuron in the third hidden layer and the neuron in the output layer is the input of the neuron in the first hidden layer is

[0112] the output of the neuron in the first hidden layer is

[0113]

[0114] the input of the neuron in the second hidden layer is

[0115]

[0116] the output of the neuron in the second hidden layer is

[0117]

[0118] the input of the neuron in the third hidden layer is

[0119]

[0120] the output of the neuron in the third hidden layer is

[0121]

[0122] the output of the neuron in the output layer is

[0123] .

[0124] Step four: constantly adjust the internal parameters of the model and the number of iterations;

[0125] The number of iterations in the set value is set to 5000 rounds, and the learning rate is set to 0.01, wherein is the gate activation function, using the sigmoid function, is the state activation function, using the tanh function. Adjust the learning rate and the number of iterations according to the set value, obtain the weight W, the loop weight A and the bias b, and adjust the corresponding hyperparameters of the input gate t, the forget gate f, the candidate gate g and the output gate j using the gate activation function, wherein the internal learning weight of the network is:

[0126]

[0127]

[0128]

[0129] The process of iteration is:

[0130]

[0131]

[0132] wherein, is the gate activation function, using the sigmoid function, that is , is the state activation function, using the tanh function;

[0133] The forget gate uses the information of and , and outputs a vector between 0 and 1 using the Sigmoid unit, and the value between 0 and 1 in the vector indicates which information in the input network model is retained and which information needs to be discarded; 0 means not to retain, and 1 means to retain all;

[0134]

[0135] The input gate is used to update the unit state, and the information of the previous hidden state and the current input information is input to the Sigmoid function first;

[0136]

[0137] The data is input into the model for model training, and two training termination indicators and are set., if any of the values of the training termination indicators and are less than a set value, then the model internal hyperparameters are readjusted for model training until both training termination indicator values meet the requirements, wherein the calculation formula of the model training termination indicator is:

[0138] wherein is the Euclidean distance, is the number of screened data, is the maximum number of iterations, is the current number of iterations.

[0139] Step five: according to the difference between the provided data and the thermal ionization particles, evaluate whether the gas fluctuation belongs to thermal ionization;

[0140] If the difference e between the output result y (collected particle data) and the thermal ionization characteristic information data compared is within the allowable error range, the particle is judged as thermal ionization, and the historical data of the particle is retained;

[0141] If the difference e between the output result y (collected particle data) and the thermal ionization characteristic information data H compared exceeds the allowable error range, the particle is not considered as thermal ionization, and the historical data of the particle is removed;

[0142] The allowable error range is

[0143]

[0144] Step six: input the screened thermal ionization historical concentration data into the established thermal ionization concentration prediction model to predict the concentration data of thermal ionization in the future 1-3 hours;

[0145] The established thermal ionization concentration prediction model is as follows:

[0146] Fitting equation:

[0147]

[0148] ;

[0149] Let be the sampling time, wherein i is 1, 2,..., 120, be the corresponding historical thermal ionization concentration, be the corresponding historical thermal ionization concentration, The fitted value of the pyroelectric ion concentration is the predicted pyroelectric ion concentration data for the next 1-3 hours;

[0150] Calculate the variance:

[0151]

[0152] If Q ≤ 1.05, proceed to step seven; otherwise, return to step two.

[0153] Step 7: Compare the predicted data with the pyroelectric ion concentration fluctuation curve at the time of the fire. If the predicted curve matches the pyroelectric ion curve at the time of the fire and also fluctuates exponentially, it is determined that there is a potential fire hazard in the future. The system sends an early warning signal to the substation inspector to remind them that a fire is about to occur. The matching of the predicted curve with the pyroelectric ion curve at the time of the fire means that the difference between the concentration of pyroelectric ions and the concentration value of pyroelectric ions at the time of the fire does not exceed 7% of the concentration value of pyroelectric ions at the time of the fire.

[0154] Table 1 shows the predicted pyroelectric ion concentration data obtained from the practical application of this method at the Jinzhou Kehong Substation:

[0155] Table 1. Predicted data on pyroelectric ion concentrations

[0156]

[0157] The concentration fluctuation curve of pyrolyzed ions during the substation fire is shown in the figure. Figure 3 As shown in the figure; a comparison chart of predicted pyroelectric ion concentration and actual pyroelectric ion concentration during a fire, as shown in the figure. Figure 4 As shown.

[0158] Calculate the variance:

[0159]

[0160] The larger the variance, the greater the deviation between the predicted result and the actual value; the smaller the variance, the closer the predicted result is to the actual value of the pyroelectric ion concentration. In this example, Q=1.02, indicating that the prediction model has high accuracy.

[0161] like Figure 5 As shown, the present invention relates to a substation fire early warning system based on pyroelectric ion detection, comprising:

[0162] The pyroelectric ion identification model building module is used to build a pyroelectric ion identification model based on a long short-term memory neural network, which includes a data acquisition layer, an input layer, a hidden layer, a network training layer, a dropout layer, a fully connected layer, a softmax layer, and an output layer.

[0163] A particulate data collection module is configured to collect particulate data in the substation according to sensors placed at various points in the substation.

[0164] A feature extraction module is configured to perform feature extraction according to the data collected by the particulate data collection module.

[0165] A parameter and iteration number adjustment module is configured to input the data into a model for model training until a training termination index value meets a requirement.

[0166] A thermal ion evaluation module is configured to evaluate whether the gas fluctuation belongs to thermal ions according to a difference between the provided data and thermal ion particulates.

[0167] A thermal ion concentration data prediction module is configured to input the screened thermal ion historical concentration data into an established thermal ion concentration prediction model to predict concentration data of thermal ions in the future 1-3 hours.

[0168] A fire hazard determination and early warning module is configured to determine that a fire will occur in the future when the prediction curve coincides with a thermal ion curve when a fire occurs and also presents an exponential form fluctuation, and to send a warning signal to a patrol officer of the substation.

[0169] A readable computer storage medium storing a computer program is provided, and the program is executed by a processor to implement the method described above.

[0170] The above is only a specific embodiment of the present application and is not used to limit the present application. For those skilled in the art, the present application can have various changes and variations. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A substation fire warning method, characterized by: Step one: build a thermal release ion identification model according to the long short-term memory neural network; Step two: collect particle data in the substation through sensors placed at various points in the substation; Step three: feature extraction on the collected data; Step four: continuously adjust the model internal parameters and iteration times; Step five: evaluate whether the gas fluctuation belongs to thermal release ion according to the difference between the provided data and the thermal release ion particle; Step six: input the selected thermal release ion historical concentration data into the established thermal release ion concentration prediction model to predict the concentration data of thermal release ion in the next 1-3 hours; Step seven: compare the predicted data with the concentration fluctuation curve of thermal release ion when a fire occurs. If the predicted curve matches the thermal release ion curve when a fire occurs and also shows exponential form fluctuation, it is determined that there is a fire hazard in the future, and the system sends a warning signal to the substation patrol personnel to remind the patrol personnel of the impending fire.

2. The substation fire warning method of claim 1, wherein: In step one, the thermal release ion identification model includes a data collection layer, an input layer, a hidden layer, a network training layer, a dropout layer, a fully connected layer, a softmax layer, and an output layer. The data collection layer is used to collect and process particle data received by sensors in the substation. The input layer is used to convert the processed data into the format of network input. The network training layer is used to train the model. The dropout layer is used to prevent network overfitting. The fully connected layer is used to extract features from the input data. The hidden layer is used to place the feature data obtained after the fully connected layer performs feature extraction. The softmax layer is used to classify the input data based on the feature extraction results and select the data with the smallest difference from the feature information. The output layer is used to output the final data obtained after classification and screening. In step two, the particle data in the substation is collected, which includes smoke particle data, thermal release ion data, and dust particle data. A group of particle data is collected every 3-5 minutes for storage, and the collection time is 2-3 hours. If there are missing data in multiple collection point data, they are filled in or repeated data are deleted. The specific process of feature extraction on the collected data in step three is as follows: For the feature information of the collected data, the particle diameter size S, light transmittance P, and particle density f are analyzed: The activation function of the first hidden layer node is a linear function, and its expression is In step four, the model internal parameters and iteration times are continuously adjusted. The specific process is as follows: adjust the learning rate and iteration times according to the set value, obtain the weight W, the loop weight A, and the bias b, and adjust the hyperparameters of the corresponding input gate t, the forget gate f, the candidate gate g, and the output gate j using the gate activation function, where the network internal learning weight is: In step five, the evaluation process of thermal release ion is as follows: The specific process of step six is as follows: Fitting equation:

3. The substation fire warning method of claim 1, wherein, Calculate the variance size:

4. The substation fire warning method of claim 1, wherein: ​ ​ The output result y is the collected particle data, and it is associated with the thermal ion characteristic information data The difference of the comparison is e: At the same time, the hidden layer added after the input layer corresponds to three characteristic information, i.e. particle diameter size S, light transmittance P and particle density size f; the first layer, the second layer and the third layer of the hidden layer respectively adopt 30-60 nodes, and a corresponding number of multiple groups of sampling point data; the output layer adopts 1 node, and corresponds to the calculated difference e; ​ The activation function of the second and third hidden layer nodes is selected as a sigmoid function, and its expression is wherein each node corresponds to a neuron, the connection weight between the neurons of the input layer and the neurons in the first hidden layer is , the connection weight between the neurons in the first hidden layer and the neurons in the second hidden layer is the same as the connection weight between the neurons in the second hidden layer and the neurons in the third hidden layer is , and the connection weight between the neurons in the third hidden layer and the neurons of the output layer is and the input of the neurons in the first hidden layer is The output of the neurons in the first hidden layer is: The input to a neuron in the second hidden layer is: The output of the neurons in the second hidden layer is: The input to the neurons in the third hidden layer is: The output of the neurons in the third hidden layer is: the output of the output layer neurons is 。 5. The substation fire warning method of claim 1, wherein: ​ The iterative process is as follows: where, is the gate activation function, using the sigmoid function, i.e. , is the state activation function, using the tanh function; Forget gate passes through and information, using a sigmoid unit to output a vector between 0 and 1, where the 0 to 1 values in the vector indicate which information to keep and which to discard in the input network model; 0 means not to keep, 1 means to keep all; The input gate is used to update the cell state, and the information of the previous hidden state and the current input information are input to the Sigmoid function first; The data is input into the model for model training, and two training termination indicators are set and If any one of the values of the training termination indicators and is less than the set value, the internal hyperparameters of the model are adjusted and the model is trained again until the values of the two training termination indicators meet the requirements, wherein the calculation formula of the model training termination indicator is: wherein is the Euclidean distance, is the number of data after screening, is the maximum number of iterations, is the current number of iterations.

6. The substation fire warning method of claim 1, wherein: ​ If the collected particle data y is compared with the thermal ion characteristic information data If the difference e is within the allowable error range, the particle is determined to be a thermal ion, and the history data of the particle is retained. If the collected particle data y is compared with the thermal ion characteristic information data If the difference e is beyond the allowable error range, the particle is determined not to be a thermal ion, and the history data of the particle is removed.

7. The substation fire warning method of claim 1, wherein: ​ ​ Let be the sampling time, where i is 1, 2,..., 120, be the corresponding historical thermal ion concentration, be the fitted value of the thermal ion concentration, i.e. the predicted future 1-3 hour thermal ion concentration data; ​ If Q < 1.05, go to Step Seven, otherwise return to Step Two.

8. The substation fire warning method of claim 1, wherein: The prediction curve in step seven matches the heat release ion curve when the fire occurs, which means that the difference between the concentration of heat release ions and the concentration value data of heat release ions when the fire occurs is not more than 7% of the concentration value of heat release ions when the fire occurs.

9. A substation fire early warning system based on thermionic emission detection, characterized in that: The method comprises the following steps: a heat release ion identification model construction module, configured to construct a heat release ion identification model comprising a data collection layer, an input layer, a hidden layer, a network training layer, a dropout layer, a full connection layer, a softmax layer, and an output layer according to a long short-term memory neural network; a particle data collection module, configured to collect particle data in a substation according to sensors placed at each point in the substation; a feature extraction module, configured to perform feature extraction according to the data collected by the particle data collection module; a parameter and iteration number adjustment module, configured to input the data into the model for model training until the training termination index value meets the requirements; a heat release ion evaluation module, configured to evaluate whether the gas fluctuation belongs to heat release ions according to the difference between the provided data and the heat release ion particles; a heat release ion concentration data prediction module, configured to input the screened heat release ion historical concentration data into the established heat release ion concentration prediction model to predict the concentration data of heat release ions in the future 1-3 hours; a fire hazard determination and early warning module, configured to determine that there will be a fire in the future when the prediction curve matches the heat release ion curve when the fire occurs and also presents an exponential form fluctuation, and send a warning signal to the patrol personnel of the substation.

10. A readable computer storage medium storing a computer program, characterized in that, The program is executed by the processor to implement the method of any one of claims 1-8.

Images