A method for predicting water inflow of sewage treatment plants based on peripheral data

Abstract

The present invention relates to the technical field of sewage treatment, and in particular to a method for predicting the water inflow of a sewage treatment plant based on peripheral data, comprising: obtaining peripheral data; the peripheral data includes environmental meteorological data, population data and enterprise data; inputting the peripheral data into a pre-trained water inflow prediction model; the pre-trained water inflow prediction model characterizes the prediction conversion relationship between the peripheral data and the water inflow prediction result; and outputting the corresponding water inflow prediction result. The method for predicting the water inflow of a sewage treatment plant in the present invention can accurately and effectively predict the water inflow of a sewage treatment plant, thereby assisting in improving the sewage management effect of the sewage treatment plant.

Description

Technical Field

[0001] The present invention relates to the technical field of sewage treatment, and in particular to a method for predicting water inflow of a sewage treatment plant based on peripheral data. Background Art

[0002] With the rapid development of the economy, the amount of urban sewage produced is a determining factor in planning the layout of the sewage system, the construction of supporting pipe networks and sewage interception systems, the scale of urban sewage treatment plants, and sewage treatment processes. In the design process of sewage treatment plants, the water quality and water volume of the designed influent often determine the scale of urban sewage treatment plants, the selection of process flows, and the investment and operating costs of the project. The actual influent water quality and water volume of most of the urban sewage treatment plants that have been built are significantly different from the design values, which seriously affects the stable and efficient operation of urban sewage treatment.

[0003] Therefore, it is necessary to predict the inflow of sewage treatment plants to provide a basis for the operation and decision-making of sewage treatment plants. For example, the Chinese patent with publication number CN106200381A discloses "A method for controlling the operation of a water plant in stages according to the amount of treated water", which includes: collecting the real-time inflow flow data of the sewage treatment plant, and calculating the current inflow chemical oxygen demand, storing the collected real-time inflow flow, inflow chemical oxygen demand data and the corresponding time points in the database; establishing a data curve for the collected data with the real-time inflow flow and inflow chemical oxygen demand data as the vertical coordinates and the time points corresponding to the real-time inflow flow data as the horizontal coordinates, and smoothing and denoising the data curve; using the neural network technology to establish a data prediction model for the smoothed and denoised data; predicting the data at the selected time to obtain the predicted value; and controlling the operation of the sewage treatment plant according to the obtained predicted value.

[0004] The method of controlling the operation of the water plant in stages in the above-mentioned existing scheme can obtain the current water inflow, and can predict the water inflow in the next time period based on the current water volume. Therefore, it is also a method for predicting the water inflow of a sewage treatment plant. The applicant found that the water inflow of a sewage treatment plant is not only related to the current data, but also related to the "peripheral data" of the location of the sewage treatment plant. For example, peripheral data such as rainfall, population, and number of enterprises will affect the water inflow of the sewage treatment plant. However, the existing method for predicting the water inflow of a sewage treatment plant only predicts the water inflow through real-time water inflow flow and inflow chemical oxygen demand, that is, it only considers the current data but not the peripheral data, so that the prediction of the water inflow is not accurate, resulting in poor sewage management effect of the sewage treatment plant. Therefore, how to provide a method for predicting the water inflow of a sewage treatment plant that can accurately and effectively predict the water inflow of a sewage treatment plant is a technical problem that needs to be solved urgently. Summary of the invention

[0005] In view of the above-mentioned deficiencies of the prior art, the technical problem to be solved by the present invention is: how to provide a sewage treatment plant inflow prediction method that can accurately and effectively predict the sewage treatment plant inflow, thereby assisting in improving the sewage management effect of the sewage treatment plant.

[0006] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0007] A method for predicting water inflow of a sewage treatment plant based on peripheral data comprises the following steps:

[0008] S01: Acquire peripheral data; the peripheral data includes environmental meteorological data, population data and enterprise data;

[0009] S02: inputting the peripheral data into a pre-trained water inflow prediction model; the pre-trained water inflow prediction model represents the prediction conversion relationship between the peripheral data and the water inflow prediction result;

[0010] S03: Output the corresponding water inflow prediction result.

[0011] Preferably, the environmental meteorological data includes any one or more of rainfall, temperature and relative humidity within a certain period of time in the future.

[0012] Preferably, the population data includes any one or more of the number of permanent urban population, the number of permanent urban population, the number of permanent rural population and the number of floating population.

[0013] Preferably, the enterprise data includes the number of each type of enterprise and its corresponding sewage generation volume.

[0014] Preferably, after obtaining the peripheral data: generate corresponding environmental meteorological database, population database and enterprise database based on the environmental meteorological data, population data and enterprise data respectively; then generate the corresponding peripheral database based on the environmental meteorological database, population database and enterprise database; finally, input the peripheral database into the water inflow prediction model to output the corresponding water inflow prediction result.

[0015] Preferably, the environmental meteorological database is represented by the following formula:

[0016] A1=a1W1+a2W2+a3W3; where: A1 represents the amount of environmental sewage generated; W1, W2, W3 represent rainfall, temperature and relative humidity respectively; a1, a2, a3 represent the coefficients corresponding to the set rainfall, temperature and relative humidity respectively;

[0017] The population database is represented by the following formula:

[0018] A2=(225×p1+225×p2+89×p3+157×M)×80%; Wherein: A2 represents the amount of domestic sewage generated; P1, P2, P3, and M represent the number of permanent urban population, the number of permanent urban population, the number of permanent rural population, and the number of floating population, respectively;

[0019] The enterprise database is represented by the following formula:

[0020] A3=b1E1+b2E2+b3E3....+b n E n ; In the formula: A3 represents the amount of industrial wastewater generated; E x represents the number of enterprises in the xth category; b x Represents the amount of sewage generated by the x-th type of enterprise; x=1,2,...n.

[0021] Preferably, the water inflow prediction model is a grey relational analysis-gated unit neural network model, and the grey relational analysis-gated unit neural network model is trained by the following steps:

[0022] S11: Establishing grey relational analysis model and gated unit neural network model respectively;

[0023] S12: connecting the output of the grey relational analysis model with the input of the gated unit neural network model to form a grey relational analysis-gated unit neural network model;

[0024] S13: training and optimizing the grey relational analysis-gated unit neural network model through the set peripheral data training set to obtain the trained grey relational analysis-gated unit neural network model.

[0025] Preferably, in step S11, establishing the grey relational analysis model includes the following steps:

[0026] S21: Generate a data matrix using the following formula and the peripheral data training set:

[0027]

[0028] Where: represents the i-th type of peripheral data, i = 1, 2, ..., α, α represents the number of categories of peripheral data;

[0029] in, Where: represents the kth type of peripheral data in the i-th type of peripheral data; γ represents the number of data types in each type of peripheral data;

[0030] S22: Generate a data matrix using the following formula and the peripheral data training set:

[0031]

[0032] Where: represents the i-th type of peripheral data, i = 1, 2, ..., α, α represents the number of categories of peripheral data;

[0033] in, Where: represents the kth type of peripheral data in the i-th type of peripheral data; γ represents the number of data types in each type of peripheral data;

[0034] S23: Determine the reference data sequence using the following formula and the peripheral data training set:

[0035] Where: represents the i-th type of reference data, i = 1, 2, ..., α; represents the kth reference data in the i-th reference data;

[0036] S24: The data matrix is ​​transformed using the following formula and a dimensionless data matrix is ​​constructed:

[0037]

[0038] S24: Calculate the correlation coefficient using the following formula:

[0039]

[0040] Where: α (k) represents the correlation coefficient between the peripheral data and the corresponding reference data; ε is the discriminant coefficient, ε∈(0,1), and the data with a correlation coefficient less than 0.7 are filtered out.

[0041] Preferably, in step S11, the established gating unit neural network model is expressed by the following formula:

[0042] zt=σ(W z ·[X t ,h t-1 ]);

[0043] rt=σ(W r ·[X t ,h t-1 ]);

[0044]

[0045]

[0046] Where: z t represents the update gate; r t Reset gate; X t Indicates the current input; ht-1 Represents the hidden layer output at time t-1; represents the summary of input and past hidden state; h t represents the hidden layer output at time t; W z , W r , All represent trainable parameter matrices; [] represents two vectors connected; * represents matrix product.

[0047] Compared with the prior art, the method for predicting the inflow of a sewage treatment plant in the present invention has the following beneficial effects:

[0048] In the present invention, the water inflow of the sewage treatment plant is predicted based on peripheral data, that is, the peripheral data such as environmental meteorological data, population data and enterprise data of the location of the sewage treatment plant are fully considered, so that the water inflow of the sewage treatment plant can be accurately and effectively predicted, thereby assisting in improving the sewage management effect of the sewage treatment plant. At the same time, the peripheral data of the present invention includes environmental meteorological data, population data and enterprise data, and these data are the peripheral data most relevant to the water inflow of the sewage treatment plant, so based on these peripheral data, the prediction accuracy of the water inflow of the sewage treatment plant can be further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to make the purpose, technical solution and advantages of the invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings, in which:

[0050] Figure 1 It is a logic block diagram of the method for predicting the water inflow of a sewage treatment plant in the embodiment;

[0051] Figure 2 This is a schematic diagram of the flow chart of the water inflow prediction model in the embodiment;

[0052] Figure 3 and Figure 4 All of them are interface diagrams of the historical database in the embodiments;

[0053] Figure 5 This is a schematic diagram of the interface during model training and testing in the embodiment;

[0054] Figure 6 This is a schematic diagram of the parameter interface for water inflow prediction in the embodiment;

[0055] Figure 7 This is a schematic diagram of the interface for predicting water inflow in the embodiment;

[0056] Figure 8 This is a schematic diagram of the interface for comparative analysis of predicted values ​​and historical actual values ​​in the embodiment. DETAILED DESCRIPTION

[0057] The following is a further detailed description through specific implementation methods:

[0058] Example:

[0059] This embodiment discloses a method for predicting the water inflow of a sewage treatment plant based on peripheral data.

[0060] like Figure 1 As shown, a method for predicting the water inflow of a sewage treatment plant based on peripheral data includes the following steps:

[0061] S01: Obtain peripheral data. Peripheral data include environmental meteorological data, population data and enterprise data. Specifically, environmental meteorological data include rainfall, temperature and relative humidity in a certain period of time in the future. Population data includes the number of permanent urban population, permanent urban population, permanent rural population and floating population. Enterprise data includes the number of various types of enterprises and their corresponding sewage generation.

[0062] S02: Inputting the peripheral data into a pre-trained water inflow prediction model, wherein the pre-trained water inflow prediction model represents a prediction conversion relationship between the peripheral data and a water inflow prediction result.

[0063] S03: Output the corresponding water inflow prediction result.

[0064] Specifically, the applicant also conducted the following research when designing the technical solution of the present invention:

[0065] The amount of rainfall has a direct impact on the amount of water entering the sewage treatment plant. When the rainfall increases, the total urban drainage volume increases. Due to incomplete rainwater and sewage diversion, leakage of sewage pipes, etc., rainwater will mix into the sewage pipes, resulting in an increase in the amount of water entering the sewage treatment plant. The effect of temperature on the amount of water entering is reflected in the effect on the activity of microorganisms in the activated sludge. Most microorganisms in sewage treatment are suitable for growth at 15°C-35°C. Within the suitable temperature range, the higher the temperature, the stronger the activity of the microorganisms, and the better the treatment effect. On the contrary, the lower the temperature, the worse the biological activity; exceeding the range inhibits the growth of microorganisms. Relative humidity (climate characteristics) also has an important impact on the amount of water entering. Relative humidity directly determines the rainfall in a certain season or period, which in turn affects the amount of water entering. This embodiment obtains the real-time and next 40-day rainfall, temperature and relative humidity of the location of the sewage treatment plant through the website of a certain regional meteorological bureau or China Weather Network.

[0066] The population size will affect the water intake of the sewage treatment plant, because the population size determines the amount of domestic sewage generated, which in turn affects the water intake. According to the 2019 China Water Resources Bulletin, my country's per capita comprehensive water consumption is 431m3, the per capita domestic water consumption (including public water) in urban areas is 225L / d, and the per capita domestic water consumption of rural residents is 89L / d; the total amount of domestic sewage generated is generally about 80% of the water consumption, so the water intake can be predicted by the population size. In this embodiment, the number of permanent urban population, permanent urban population and permanent rural population in a certain area can be checked through the statistical information network of a certain area, and then the number of tourists received during major holidays each year can be counted as the number of floating population through the official website of the government of a certain area.

[0067] The number of enterprises will affect the water inflow of the sewage treatment plant, because the number of enterprises directly affects the industrial water consumption, which in turn affects the water inflow of the sewage treatment plant. Therefore, the corresponding water inflow can be predicted by the number of various types of enterprises. This embodiment consults the number of enterprises of different categories above a certain scale and the corresponding water consumption through the regional statistical information network and the Water Conservancy Bureau; then consults the sewage discharge coefficient (the ratio of sewage discharge to water consumption) of relevant enterprises according to the "Manual of Pollution Production and Discharge Coefficients", and determines the corresponding sewage generation of various types of enterprises by linear analysis.

[0068] In the present invention, the water inflow of the sewage treatment plant is predicted based on peripheral data, that is, the peripheral data such as environmental meteorological data, population data and enterprise data of the location of the sewage treatment plant are fully considered, so that the water inflow of the sewage treatment plant can be accurately and effectively predicted, thereby assisting in improving the sewage management effect of the sewage treatment plant. At the same time, the peripheral data of the present invention includes environmental meteorological data, population data and enterprise data, and these data are the peripheral data most relevant to the water inflow of the sewage treatment plant, so based on these peripheral data, the prediction accuracy of the water inflow of the sewage treatment plant can be further improved.

[0069] During the specific implementation process, after obtaining the peripheral data: the corresponding environmental meteorological database, population database and enterprise database are generated based on the environmental meteorological data, population data and enterprise data respectively; then the corresponding peripheral database is generated based on the environmental meteorological database, population database and enterprise database; finally, the peripheral database is input into the water inflow prediction model to output the corresponding water inflow prediction result.

[0070] Specifically, the environmental meteorological database is expressed by the following formula:

[0071] A1=a1W1+a2W2+a3W3; where: A1 represents the amount of environmental sewage generated; W1, W2, W3 represent rainfall, temperature and relative humidity respectively; a1, a2, a3 represent the coefficients corresponding to the set rainfall, temperature and relative humidity respectively;

[0072] The population database is represented by the following formula:

[0073] A2=(225×p1+225×p2+89×p3+157×M)×80%; Wherein: A2 represents the amount of domestic sewage generated; P1, P2, P3, and M represent the number of permanent urban population, the number of permanent urban population, the number of permanent rural population, and the number of floating population, respectively;

[0074] The enterprise database is represented by the following formula:

[0075] A3=b1E1+b2E2+b3E3....+b n E n ; In the formula: A3 represents the amount of industrial wastewater generated; E x represents the number of enterprises in the xth category; b x Represents the amount of sewage generated by the x-th type of enterprise; x=1,2,...n.

[0076] In the present invention, environmental meteorological data, population data and enterprise data are sorted and classified by correspondingly generating a database, so that various peripheral data will not affect each other, which is helpful to assist in improving the prediction accuracy of the water inflow of the sewage treatment plant.

[0077] In the specific implementation process, Figure 2 As shown, the water inflow prediction model is a grey relational analysis-gated unit neural network model (GRA-GRU model), and the grey relational analysis-gated unit neural network model is trained by the following steps:

[0078] S11: Establishing grey relational analysis model and gated unit neural network model respectively;

[0079] S12: connecting the output of the grey relational analysis model with the input of the gated unit neural network model to form a grey relational analysis-gated unit neural network model;

[0080] S13: training and optimizing the grey relational analysis-gated unit neural network model through the set peripheral data training set to obtain the trained grey relational analysis-gated unit neural network model.

[0081] Specifically, the establishment of the grey relational analysis model includes the following steps:

[0082] S21: Generate a data matrix using the following formula and the peripheral data training set:

[0083]

[0084] Where: represents the i-th type of peripheral data, i = 1, 2, ..., α, α represents the number of categories of peripheral data;

[0085] in, Where: represents the kth type of peripheral data in the i-th type of peripheral data; γ represents the number of data types in each type of peripheral data;

[0086] S22: Generate a data matrix using the following formula and the peripheral data training set:

[0087]

[0088] Where: represents the i-th type of peripheral data, i = 1, 2, ..., α, α represents the number of categories of peripheral data;

[0089] in, Where: represents the kth type of peripheral data in the i-th type of peripheral data; γ represents the number of data types in each type of peripheral data;

[0090] S23: Determine the reference data sequence using the following formula and the peripheral data training set:

[0091] Where: represents the i-th type of reference data, i = 1, 2, ..., α; represents the kth reference data in the i-th reference data;

[0092] S24: The data matrix is ​​transformed using the following formula and a dimensionless data matrix is ​​constructed:

[0093]

[0094] S24: Calculate the correlation coefficient using the following formula:

[0095]

[0096] Where: α (k) represents the correlation coefficient between the peripheral data and the corresponding reference data; ε is the discriminant coefficient, ε∈(0,1), and the data with a correlation coefficient less than 0.7 are filtered out.

[0097] Specifically, the established gating unit neural network model is expressed by the following formula:

[0098] z t =σ(W z ·[X t ,h t-1 ]);

[0099] r t=σ(W r ·[X t ,h t-1 ]);

[0100]

[0101]

[0102] Where: z t represents the update gate; r t Reset gate; X t Indicates the current input; h t-1 Represents the hidden layer output at time t-1; represents the summary of input and past hidden state; h t represents the hidden layer output at time t; W z , W r , All represent trainable parameter matrices; [] represents two vectors connected; * represents matrix product.

[0103] In the present invention, the grey relational analysis-gated unit neural network model (GRA-GRU model) can accurately and effectively predict the water inflow of the sewage treatment plant, which is helpful to assist in improving the sewage management effect of the sewage treatment plant. At the same time, by establishing and training the model in the above manner, a grey relational analysis-gated unit neural network model with extremely high accuracy can be obtained, which can also improve the prediction accuracy of the water inflow of the sewage treatment plant.

[0104] In order to verify the feasibility of the method for predicting the water inflow of a sewage treatment plant of the present invention, this embodiment takes a sewage treatment plant in a certain city as an example and conducts the following experiments:

[0105] The historical data of enterprises, population, meteorology and sewage treatment plants in the coverage area of ​​the city’s sewage treatment plant are collected, among which the historical data of the sewage treatment plant include treatment volume, total power consumption, average power consumption, etc.

[0106] Establish a historical database as a peripheral data training set. The historical data is as follows: Figure 3 and Figure 4 shown.

[0107] The established database was summarized and correlation analysis was performed.

[0108] Import the data into the Grey Relational Analysis-Gated Unit Neural Network Model (GRA-GRU model) for model training and optimization (testing). The training interface and results are as follows: Figure 5 shown.

[0109] According to the collected environmental meteorological database, population database and enterprise database, the enterprise, population and meteorological data are input into the grey correlation analysis-gated unit neural network model to predict the water inflow. The prediction parameter interface is as follows: Figure 6 As shown, the water inflow prediction interface is as follows Figure 7 shown.

[0110] Compare and analyze the predicted water inflow value with the historical actual value. The comparison analysis interface is as follows: Figure 8 shown.

[0111] The water inflow prediction accuracy of the present invention reaches 95.52%, and the prediction result meets the standard. Therefore, the sewage treatment plant water inflow prediction method of the present invention is feasible.

[0112] The above is only an embodiment of the present invention. The common sense such as the known specific structure and characteristics in the scheme is not described in detail here. The ordinary technicians in the relevant field know all the common technical knowledge in the technical field of the invention before the application date or priority date, can obtain all the existing technologies in the field, and have the ability to apply the conventional experimental means before that date. The ordinary technicians in the relevant field can improve and implement this scheme in combination with their own abilities under the enlightenment given by this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the relevant field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several deformations and improvements can be made, which should also be regarded as the protection scope of the present invention, which will not affect the effect of the implementation of the present invention and the practicality of the patent. The protection scope required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A method for predicting water inflow of a sewage treatment plant based on peripheral data, characterized in that: The following steps are involved: S01: Acquire peripheral data; the peripheral data includes environmental meteorological data, population data and enterprise data; S02: inputting the peripheral data into a pre-trained water inflow prediction model; the pre-trained water inflow prediction model represents the prediction conversion relationship between the peripheral data and the water inflow prediction result; The water inflow prediction model is a grey relational analysis-gated unit neural network model, and the grey relational analysis-gated unit neural network model is trained by the following steps: S11: Establishing grey relational analysis model and gated unit neural network model respectively; S12: connecting the output of the grey relational analysis model with the input of the gated unit neural network model to form a grey relational analysis-gated unit neural network model; S13: training and optimizing the grey relational analysis-gated unit neural network model through the set peripheral data training set to obtain a trained grey relational analysis-gated unit neural network model; In step S11, the establishment of the grey relational analysis model includes the following steps: S21: Generate a data matrix using the following formula and the peripheral data training set: Where: represents the i-th type of peripheral data, i = 1, 2, ..., α, α represents the number of categories of peripheral data; in, Where: represents the kth type of peripheral data in the i-th type of peripheral data; γ represents the number of data types in each type of peripheral data; S22: Generate a data matrix using the following formula and the peripheral data training set: Where: represents the i-th type of peripheral data, i = 1, 2, ..., α, α represents the number of categories of peripheral data; in, Where: represents the kth type of peripheral data in the i-th type of peripheral data; γ represents the number of data types in each type of peripheral data; S23: Determine the reference data sequence using the following formula and the peripheral data training set: Where: represents the i-th type of reference data, i = 1, 2, ..., α; represents the kth reference data in the i-th reference data; S24: The data matrix is ​​transformed using the following formula and a dimensionless data matrix is ​​constructed: S24: Calculate the correlation coefficient using the following formula: Where: α (k) represents the correlation coefficient between the peripheral data and the corresponding reference data; ε is the discriminant coefficient, ε∈(0,1), and the data with a correlation coefficient less than 0.7 are filtered out; S03: Output the corresponding water inflow prediction result.

2. The method for predicting water inflow of a sewage treatment plant based on peripheral data according to claim 1, characterized in that: The environmental meteorological data includes any one or more of rainfall, temperature and relative humidity within a certain period of time in the future.

3. The method for predicting water inflow of a sewage treatment plant based on peripheral data according to claim 1, characterized in that: The population data includes any one or more of the number of permanent urban population, the number of permanent urban population, the number of permanent rural population and the number of floating population.

4. The method for predicting water inflow of a sewage treatment plant based on peripheral data according to claim 1, characterized in that: The enterprise data includes the number of various types of enterprises and their corresponding sewage generation.

5. The method for predicting water inflow of a sewage treatment plant based on peripheral data according to claim 1, characterized in that: After obtaining the peripheral data: generate corresponding environmental meteorological database, population database and enterprise database based on environmental meteorological data, population data and enterprise data respectively; then generate corresponding peripheral database based on environmental meteorological database, population database and enterprise database; finally, input the peripheral database into the water inflow prediction model to output the corresponding water inflow prediction result.

6. The method for predicting water inflow of a sewage treatment plant based on peripheral data according to claim 1, characterized in that: The environmental meteorological database is expressed by the following formula: A1=a1W1+a2W2+a3W3; where: A1 represents the amount of environmental sewage generated; W1, W2, W3 represent rainfall, temperature and relative humidity respectively; a1, a2, a3 represent the coefficients corresponding to the set rainfall, temperature and relative humidity respectively; The population database is represented by the following formula: A2=(225×p1+225×p2+89×p3+157×M)×80%; Wherein: A2 represents the amount of domestic sewage generated; P1, P2, P3, and M represent the number of permanent urban population, the number of permanent urban population, the number of permanent rural population, and the number of floating population, respectively; The enterprise database is represented by the following formula: A3=b1E1+b2E2+b3E3....+b n E n ; In the formula: A3 represents the amount of industrial wastewater generated; E x represents the number of enterprises in the xth category; b x Represents the amount of sewage generated by the x-th type of enterprise; x=1,2,...n.

7. The method for predicting the inflow of a sewage treatment plant based on peripheral data according to claim 1, characterized in that: In step S11, the established gating unit neural network model is expressed by the following formula: z t =σ(W z ·[X t ,h t-1 ]); r t =σ(W r ·[X t ,h t-1 ]); Where: z t represents the update gate; r t Reset gate; X t Indicates the current input; h t-1 Represents the hidden layer output at time t-1; represents the summary of input and past hidden state; h t represents the hidden layer output at time t; W z , W r , All represent trainable parameter matrices; [ ] indicates that two vectors are connected; * indicates matrix product.

Images